Merge branch 'release/v2.1' into develop

README.md

@@ -27,16 +27,16 @@ English | [简体中文](README_ch.md)
 
 
 ## Introduction and Features
-- PaddleHub aims to provide developers with rich, high-quality, and directly usable pre-trained models.
-- **Abundant Pre-trained Models**: 300+ pre-trained models covering the 5 major categories including Image, Text, Audio, Video, and Industrial application. All of them are free for download and offline usage.
-- **No need for deep learning background**: you can use AI models quickly and enjoy the dividends of the artificial intelligence era.
-- **Quick Model Prediction**: Model prediction can be realized through a few lines of scripts to quickly experience the model effect.
+- **PaddleHub** aims to provide developers with rich, high-quality, and directly usable pre-trained models.
+- **Abundant Pre-trained Models**: 300+ pre-trained models cover the 5 major categories, including Image, Text, Audio, Video, and Industrial application. All of them are free for download and offline usage.
+- **No Need for Deep Learning Background**: you can use AI models quickly and enjoy the dividends of the artificial intelligence era.
+- **Quick Model Prediction**: model prediction can be realized through a few lines of scripts to quickly experience the model effect.
 - **Model As Service**: one-line command to build deep learning model API service deployment capabilities.
-- **Easy-to-use Transfer Learning**: few lines of code to complete the transfer-learning task such as image classification and text classification based on high quality pre-trained models.
+- **Easy-to-use Transfer Learning**: few lines of codes to complete the transfer-learning task such as image classification and text classification based on high quality pre-trained models.
 - **Cross-platform**: support Linux, Windows, MacOS and other operating systems.
 
 ### Recent updates
-- **2021.05.12**，Add  an open-domain dialogue system, i.e., [plato-mini](https://www.paddlepaddle.org.cn/hubdetail?name=plato-mini&en_category=TextGeneration), to make it easy to build a chatbot in wechat with the help of the wechaty, [See Demo](https://github.com/KPatr1ck/paddlehub-wechaty-demo)
+- **2021.05.12:** Add an open-domain dialogue system, i.e., [plato-mini](https://www.paddlepaddle.org.cn/hubdetail?name=plato-mini&en_category=TextGeneration), to make it easy to build a chatbot in wechat with the help of the wechaty, [See Demo](https://github.com/KPatr1ck/paddlehub-wechaty-demo)
 - **2021.04.27:** The v2.1.0 version is released. [1] Add supports for five new models, including two high-precision semantic segmentation models based on VOC dataset and three voice classification models. [2] Enforce the transfer learning capabilities for image semantic segmentation, text semantic matching and voice classification on related datasets. [3] Add the export function APIs for two kinds of model formats, i.,e, ONNX and PaddleInference. [4] Add the support for [BentoML](https://github.com/bentoml/BentoML/), which is a cloud native framework for serving deployment. Users can easily serve pre-trained models from PaddleHub by following the [Tutorial notebooks](https://github.com/PaddlePaddle/PaddleHub/blob/release/v2.1/demo/serving/bentoml/cloud-native-model-serving-with-bentoml.ipynb). Also, see this announcement and [Release note](https://github.com/bentoml/BentoML/releases/tag/v0.12.1) from BentoML. (Many thanks to @[parano](https://github.com/parano) @[cqvu](https://github.com/cqvu) @[deehrlic](https://github.com/deehrlic) for contributing this feature in PaddleHub). [5] The total number of pre-trained models reaches **【300】**.
 - **2021.02.18:** The v2.0.0 version is released, making model development and debugging easier, and the finetune task is more flexible and easy to use.The ability to transfer learning for visual tasks is fully upgraded, supporting various tasks such as image classification, image coloring, and style transfer; Transformer models such as BERT, ERNIE, and RoBERTa are upgraded to dynamic graphs, supporting Fine-Tune capabilities for text classification and sequence labeling; Optimize the Serving capability, support multi-card prediction, automatic load balancing, and greatly improve performance; the new automatic data enhancement capability Auto Augment can efficiently search for data enhancement strategy combinations suitable for data sets. 61 new word vector models were added, including 51 Chinese models and 10 English models; add 4 image segmentation models, 2 depth models, 7 image generation models, and 3 text generation models, the total number of pre-trained models reaches **【274】**.
 - [【more】](./docs/docs_en/release.md)
@@ -50,7 +50,7 @@ English | [简体中文](README_ch.md)
 <img src="./docs/imgs/Readme_Related/Image_all.gif"  width = "530" height = "400" />
 </div>
 
-- Many thanks to CopyRight@[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)、[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)、[PaddleGAN](https://github.com/PaddlePaddle/PaddleGAN)、[AnimeGAN](https://github.com/TachibanaYoshino/AnimeGANv2)、[openpose](https://github.com/CMU-Perceptual-Computing-Lab/openpose)、[PaddleSeg](https://github.com/PaddlePaddle/PaddleSeg)、[Zhengxia Zou](https://github.com/jiupinjia/SkyAR) for the pre-trained models, you can try to train your models with them.
+- Many thanks to CopyRight@[PaddleOCR](https://github.com/PaddlePaddle/PaddleOCR)、[PaddleDetection](https://github.com/PaddlePaddle/PaddleDetection)、[PaddleGAN](https://github.com/PaddlePaddle/PaddleGAN)、[AnimeGAN](https://github.com/TachibanaYoshino/AnimeGANv2)、[openpose](https://github.com/CMU-Perceptual-Computing-Lab/openpose)、[PaddleSeg](https://github.com/PaddlePaddle/PaddleSeg)、[Zhengxia Zou](https://github.com/jiupinjia/SkyAR)、[PaddleClas](https://github.com/PaddlePaddle/PaddleClas) for the pre-trained models, you can try to train your models with them.
 
 
 ### **Natural Language Processing (129 models)**
@@ -118,14 +118,21 @@ If you have any questions during the use of the model, you can join the official
 <div align="center">
 <img src="./docs/imgs/joinus.PNG"  width = "200" height = "200" />
 </div>  
-If you fail to scan the code, please add WeChat 15704308458 and note "Hub", the operating class will invite you to join the group.
+please add WeChat above and send "Hub" to the robot, the robot will invite you to join the group automatically.
 
 <a name="QuickStart"></a>
 ## QuickStart
 
 ```python
-!pip install --upgrade paddlepaddle -i https://mirror.baidu.com/pypi/simple
-!pip install --upgrade paddlehub -i https://mirror.baidu.com/pypi/simple
+# install paddlepaddle with gpu
+# !pip install --upgrade paddlepaddle-gpu
+
+# or install paddlepaddle with cpu
+!pip install --upgrade paddlepaddle
+
+# install paddlehub
+!pip install --upgrade paddlehub
+
 
 import paddlehub as hub
 
@@ -136,6 +143,7 @@ results = lac.cut(text=test_text, use_gpu=False, batch_size=1, return_tag=True)
 print(results)
 #{'word': ['今天', '是', '个', '好天气', '。'], 'tag': ['TIME', 'v', 'q', 'n', 'w']}
 ```
+More API for transfer learning, please refer [Tutorial](https://paddlehub.readthedocs.io/en/release-v2.1/transfer_learning_index.html)
 
 <a name="License"></a>
 ## License
@@ -194,3 +202,4 @@ We welcome you to contribute code to PaddleHub, and thank you for your feedback.
 * Many thanks to [BurrowsWang](https://github.com/BurrowsWang) for fixing Markdown table display problem
 * Many thanks to [huqi](https://github.com/hu-qi) for fixing readme typo
 * Many thanks to [parano](https://github.com/parano) [cqvu](https://github.com/cqvu) [deehrlic](https://github.com/deehrlic) for contributing this feature in PaddleHub
+* Many thanks to [paopjian](https://github.com/paopjian) for correcting the wrong website address [#1424](https://github.com/PaddlePaddle/PaddleHub/issues/1424)

README_ch.md

@@ -29,11 +29,11 @@
 
 
 ## 简介与特性
-- PaddleHub旨在为开发者提供丰富的、高质量的、直接可用的预训练模型。
-- **【模型种类丰富】**: 涵盖CV、NLP、Audio、Video、工业应用主流五大品类的 300+ 预训练模型，全部开源下载，离线可运行。
-- **【超低使用门槛】**：无需深度学习背景、无需数据与训练过程，可快速使用AI模型，
-- **【一键模型快速预测】**：通过一行命令行或者极简的Python API实现模型调用，可快速体验模型效果。
-- **【一键模型转服务化】**：一行命令，搭建深度学习模型API服务化部署能力。
+- PaddleHub旨在为开发者提供丰富的、高质量的、直接可用的预训练模型
+- **【模型种类丰富】**: 涵盖CV、NLP、Audio、Video、工业应用主流五大品类的 300+ 预训练模型，全部开源下载，离线可运行
+- **【超低使用门槛】**：无需深度学习背景、无需数据与训练过程，可快速使用AI模型
+- **【一键模型快速预测】**：通过一行命令行或者极简的Python API实现模型调用，可快速体验模型效果
+- **【一键模型转服务化】**：一行命令，搭建深度学习模型API服务化部署能力
 - **【十行代码迁移学习】**：十行代码完成图片分类、文本分类的迁移学习任务
 - **【跨平台兼容性】**：可运行于Linux、Windows、MacOS等多种操作系统
 
@@ -124,7 +124,7 @@
 <div align="center">
 <img src="./docs/imgs/joinus.PNG"  width = "200" height = "200" />
 </div>  
-如扫码失败，请添加微信15704308458，并备注“Hub”，运营同学会邀请您入群。  
+扫码备注"Hub"加好友之后，再发送“Hub”，会自动邀请您入群。  
 
 <div id="QuickStart">
 
@@ -149,6 +149,8 @@ print(results)
 ！hub serving start -m lac
 ```
 
+ 更多迁移学习能力可以参考[教程文档](https://paddlehub.readthedocs.io/zh_CN/release-v2.1/transfer_learning_index.html)
+
 
 
 
@@ -208,3 +210,4 @@ print(results)
 * 非常感谢[BurrowsWang](https://github.com/BurrowsWang)修复Markdown表格显示问题
 * 非常感谢[huqi](https://github.com/hu-qi)修复了readme中的错别字
 * 非常感谢[parano](https://github.com/parano)、[cqvu](https://github.com/cqvu)、[deehrlic](https://github.com/deehrlic)三位的贡献与支持
+* 非常感谢[paopjian](https://github.com/paopjian)修改了中文readme模型搜索指向的的网站地址错误[#1424](https://github.com/PaddlePaddle/PaddleHub/issues/1424)

demo/README.md

+### PaddleHub Office Website：https://www.paddlepaddle.org.cn/hub
+### PaddleHub Module Searching：https://www.paddlepaddle.org.cn/hublist

demo/text_classification/predict.py

@@ -28,6 +28,6 @@ if __name__ == '__main__':
         task='seq-cls',
         load_checkpoint='./test_ernie_text_cls/best_model/model.pdparams',
         label_map=label_map)
-    results = model.predict(data, max_seq_len=50, batch_size=1, use_gpu=False)
+    results, probs = model.predict(data, max_seq_len=50, batch_size=1, use_gpu=False, return_prob=True)
     for idx, text in enumerate(data):
         print('Data: {} \t Lable: {}'.format(text[0], results[idx]))

docs/docs_ch/get_start/installation.rst

@@ -21,7 +21,7 @@
 安装命令
 ========================
 
-在安装PaddleHub之前，请先安装PaddlePaddle深度学习框架，更多安装说明请查阅`飞桨快速安装 <https://www.paddlepaddle.org.cn/install/quick>`_.
+在安装PaddleHub之前，请先安装PaddlePaddle深度学习框架，更多安装说明请查阅`飞桨快速安装 <https://www.paddlepaddle.org.cn/install/quick>`
 
 .. code-block:: shell
 
@@ -30,6 +30,7 @@
 除上述依赖外，PaddleHub的预训练模型和预置数据集需要连接服务端进行下载，请确保机器可以正常访问网络。若本地已存在相关的数据集和预训练模型，则可以离线运行PaddleHub。
 
 .. note::
+
     使用PaddleHub下载数据集、预训练模型等，要求机器可以访问外网。可以使用`server_check()`可以检查本地与远端PaddleHub-Server的连接状态，使用方法如下：
 
 .. code-block:: Python
@@ -37,4 +38,4 @@
     import paddlehub
     paddlehub.server_check()
     # 如果可以连接远端PaddleHub-Server，则显示Request Hub-Server successfully。
-    # 如果无法连接远端PaddleHub-Server，则显示Request Hub-Server unsuccessfully。
\ No newline at end of file
+    # 如果无法连接远端PaddleHub-Server，则显示Request Hub-Server unsuccessfully。

docs/docs_ch/get_start/python_use_hub.rst

@@ -32,7 +32,7 @@ PaddleHub采用模型即软件的设计理念，所有的预训练模型与Pytho
     # module = hub.Module(name="humanseg_lite", version="1.1.1")
     module = hub.Module(name="humanseg_lite")
 
-    res = module.segmentation(
+    res = module.segment(
         paths = ["./test_image.jpg"], 
         visualization=True, 
         output_dir='humanseg_output')
@@ -131,4 +131,4 @@ PaddleHub采用模型即软件的设计理念，所有的预训练模型与Pytho
 
 ----------------
 
-    [{'text': '味道不错，确实不算太辣，适合不能吃辣的人。就在长江边上，抬头就能看到长江的风景。鸭肠、黄鳝都比较新鲜。', 'sentiment_label': 1, 'sentiment_key': 'positive', 'positive_probs': 0.9771, 'negative_probs': 0.0229}]
\ No newline at end of file
+    [{'text': '味道不错，确实不算太辣，适合不能吃辣的人。就在长江边上，抬头就能看到长江的风景。鸭肠、黄鳝都比较新鲜。', 'sentiment_label': 1, 'sentiment_key': 'positive', 'positive_probs': 0.9771, 'negative_probs': 0.0229}]

modules/audio/audio_classification/PANNs/cnn10/module.py

@@ -31,7 +31,7 @@ from paddlehub.utils.log import logger
     name="panns_cnn10",
     version="1.0.0",
     summary="",
-    author="Baidu",
+    author="paddlepaddle",
     author_email="",
     type="audio/sound_classification",
     meta=AudioClassifierModule)

modules/audio/audio_classification/PANNs/cnn14/module.py

@@ -31,7 +31,7 @@ from paddlehub.utils.log import logger
     name="panns_cnn14",
     version="1.0.0",
     summary="",
-    author="Baidu",
+    author="paddlepaddle",
     author_email="",
     type="audio/sound_classification",
     meta=AudioClassifierModule)

modules/audio/audio_classification/PANNs/cnn6/module.py

@@ -31,7 +31,7 @@ from paddlehub.utils.log import logger
     name="panns_cnn6",
     version="1.0.0",
     summary="",
-    author="Baidu",
+    author="paddlepaddle",
     author_email="",
     type="audio/sound_classification",
     meta=AudioClassifierModule)

modules/audio/voice_cloning/lstm_tacotron2/README.md

+```shell
+$ hub install lstm_tacotron2==1.0.0
+```
+
+## 概述
+
+声音克隆是指使用特定的音色，结合文字的读音合成音频，使得合成后的音频具有目标说话人的特征，从而达到克隆的目的。
+
+在训练语音克隆模型时，目标音色作为Speaker Encoder的输入，模型会提取这段语音的说话人特征（音色）作为Speaker Embedding。接着，在训练模型重新合成此类音色的语音时，除了输入的目标文本外，说话人的特征也将成为额外条件加入模型的训练。
+
+在预测时，选取一段新的目标音色作为Speaker Encoder的输入，并提取其说话人特征，最终实现输入为一段文本和一段目标音色，模型生成目标音色说出此段文本的语音片段。
+
+![](https://ai-studio-static-online.cdn.bcebos.com/982ab955b87244d3bae3b003aff8e28d9ec159ff0d6246a79757339076dfe7d4)
+
+`lstm_tacotron2`是一个支持中文的语音克隆模型，分别使用了LSTMSpeakerEncoder、Tacotron2和WaveFlow模型分别用于语音特征提取、目标音频特征合成和语音波形转换。
+
+关于模型的详请可参考[Parakeet](https://github.com/PaddlePaddle/Parakeet/tree/release/v0.3/parakeet/models)。
+
+
+## API
+
+```python
+def __init__(speaker_audio: str = None,
+             output_dir: str = './')
+```
+初始化module，可配置模型的目标音色的音频文件和输出的路径。
+
+**参数**
+- `speaker_audio`(str): 目标说话人语音音频文件(*.wav)的路径，默认为None(使用默认的女声作为目标音色)。
+- `output_dir`(str): 合成音频的输出文件，默认为当前目录。
+
+
+```python
+def get_speaker_embedding()
+```
+获取模型的目标说话人特征。
+
+**返回**
+* `results`(numpy.ndarray): 长度为256的numpy数组，代表目标说话人的特征。
+
+```python
+def set_speaker_embedding(speaker_audio: str)
+```
+设置模型的目标说话人特征。
+
+**参数**
+- `speaker_audio`(str): 必填，目标说话人语音音频文件(*.wav)的路径。
+
+```python
+def generate(data: List[str], batch_size: int = 1, use_gpu: bool = False):
+```
+根据输入文字，合成目标说话人的语音音频文件。
+
+**参数**
+- `data`(List[str]): 必填，目标音频的内容文本列表，目前只支持中文，不支持添加标点符号。
+- `batch_size`(int): 可选，模型合成语音时的batch_size，默认为1。
+- `use_gpu`(bool): 是否使用gpu执行计算，默认为False。
+
+
+**代码示例**
+
+```python
+import paddlehub as hub
+
+model = hub.Module(name='lstm_tacotron2', output_dir='./', speaker_audio='/data/man.wav')  # 指定目标音色音频文件
+texts = [
+    '语音的表现形式在未来将变得越来越重要$',
+    '今天的天气怎么样$',  ]
+wavs = model.generate(texts, use_gpu=True)
+
+for text, wav in zip(texts, wavs):
+    print('='*30)
+    print(f'Text: {text}')
+    print(f'Wav: {wav}')
+```
+
+输出
+```
+==============================
+Text: 语音的表现形式在未来将变得越来越重要$
+Wav: /data/1.wav
+==============================
+Text: 今天的天气怎么样$
+Wav: /data/2.wav
+```
+
+
+## 查看代码
+
+https://github.com/PaddlePaddle/Parakeet
+
+## 依赖
+
+paddlepaddle >= 2.0.0
+
+paddlehub >= 2.1.0
+
+## 更新历史
+
+* 1.0.0
+
+  初始发布

modules/audio/voice_cloning/lstm_tacotron2/audio_processor.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from pathlib import Path
+from warnings import warn
+import struct
+
+from scipy.ndimage.morphology import binary_dilation
+import numpy as np
+import librosa
+
+try:
+    import webrtcvad
+except ModuleNotFoundError:
+    warn("Unable to import 'webrtcvad'." "This package enables noise removal and is recommended.")
+    webrtcvad = None
+
+INT16_MAX = (2**15) - 1
+
+
+def normalize_volume(wav, target_dBFS, increase_only=False, decrease_only=False):
+    # this function implements Loudness normalization, instead of peak
+    # normalization, See https://en.wikipedia.org/wiki/Audio_normalization
+    # dBFS: Decibels relative to full scale
+    # See https://en.wikipedia.org/wiki/DBFS for more details
+    # for 16Bit PCM audio, minimal level is -96dB
+    # compute the mean dBFS and adjust to target dBFS, with by increasing
+    # or decreasing
+    if increase_only and decrease_only:
+        raise ValueError("Both increase only and decrease only are set")
+    dBFS_change = target_dBFS - 10 * np.log10(np.mean(wav**2))
+    if ((dBFS_change < 0 and increase_only) or (dBFS_change > 0 and decrease_only)):
+        return wav
+    gain = 10**(dBFS_change / 20)
+    return wav * gain
+
+
+def trim_long_silences(wav, vad_window_length: int, vad_moving_average_width: int, vad_max_silence_length: int,
+                       sampling_rate: int):
+    """
+    Ensures that segments without voice in the waveform remain no longer than a
+    threshold determined by the VAD parameters in params.py.
+
+    :param wav: the raw waveform as a numpy array of floats
+    :return: the same waveform with silences trimmed away (length <= original wav length)
+    """
+    # Compute the voice detection window size
+    samples_per_window = (vad_window_length * sampling_rate) // 1000
+
+    # Trim the end of the audio to have a multiple of the window size
+    wav = wav[:len(wav) - (len(wav) % samples_per_window)]
+
+    # Convert the float waveform to 16-bit mono PCM
+    pcm_wave = struct.pack("%dh" % len(wav), *(np.round(wav * INT16_MAX)).astype(np.int16))
+
+    # Perform voice activation detection
+    voice_flags = []
+    vad = webrtcvad.Vad(mode=3)
+    for window_start in range(0, len(wav), samples_per_window):
+        window_end = window_start + samples_per_window
+        voice_flags.append(vad.is_speech(pcm_wave[window_start * 2:window_end * 2], sample_rate=sampling_rate))
+    voice_flags = np.array(voice_flags)
+
+    # Smooth the voice detection with a moving average
+    def moving_average(array, width):
+        array_padded = np.concatenate((np.zeros((width - 1) // 2), array, np.zeros(width // 2)))
+        ret = np.cumsum(array_padded, dtype=float)
+        ret[width:] = ret[width:] - ret[:-width]
+        return ret[width - 1:] / width
+
+    audio_mask = moving_average(voice_flags, vad_moving_average_width)
+    audio_mask = np.round(audio_mask).astype(np.bool)
+
+    # Dilate the voiced regions
+    audio_mask = binary_dilation(audio_mask, np.ones(vad_max_silence_length + 1))
+    audio_mask = np.repeat(audio_mask, samples_per_window)
+
+    return wav[audio_mask]
+
+
+def compute_partial_slices(n_samples: int,
+                           partial_utterance_n_frames: int,
+                           hop_length: int,
+                           min_pad_coverage: float = 0.75,
+                           overlap: float = 0.5):
+    """
+    Computes where to split an utterance waveform and its corresponding mel spectrogram to obtain
+    partial utterances of <partial_utterance_n_frames> each. Both the waveform and the mel
+    spectrogram slices are returned, so as to make each partial utterance waveform correspond to
+    its spectrogram. This function assumes that the mel spectrogram parameters used are those
+    defined in params_data.py.
+
+    The returned ranges may be indexing further than the length of the waveform. It is
+    recommended that you pad the waveform with zeros up to wave_slices[-1].stop.
+
+    :param n_samples: the number of samples in the waveform
+    :param partial_utterance_n_frames: the number of mel spectrogram frames in each partial
+    utterance
+    :param min_pad_coverage: when reaching the last partial utterance, it may or may not have
+    enough frames. If at least <min_pad_coverage> of <partial_utterance_n_frames> are present,
+    then the last partial utterance will be considered, as if we padded the audio. Otherwise,
+    it will be discarded, as if we trimmed the audio. If there aren't enough frames for 1 partial
+    utterance, this parameter is ignored so that the function always returns at least 1 slice.
+    :param overlap: by how much the partial utterance should overlap. If set to 0, the partial
+    utterances are entirely disjoint.
+    :return: the waveform slices and mel spectrogram slices as lists of array slices. Index
+    respectively the waveform and the mel spectrogram with these slices to obtain the partial
+    utterances.
+    """
+    assert 0 <= overlap < 1
+    assert 0 < min_pad_coverage <= 1
+
+    # librosa's function to compute num_frames from num_samples
+    n_frames = int(np.ceil((n_samples + 1) / hop_length))
+    # frame shift between ajacent partials
+    frame_step = max(1, int(np.round(partial_utterance_n_frames * (1 - overlap))))
+
+    # Compute the slices
+    wav_slices, mel_slices = [], []
+    steps = max(1, n_frames - partial_utterance_n_frames + frame_step + 1)
+    for i in range(0, steps, frame_step):
+        mel_range = np.array([i, i + partial_utterance_n_frames])
+        wav_range = mel_range * hop_length
+        mel_slices.append(slice(*mel_range))
+        wav_slices.append(slice(*wav_range))
+
+    # Evaluate whether extra padding is warranted or not
+    last_wav_range = wav_slices[-1]
+    coverage = (n_samples - last_wav_range.start) / (last_wav_range.stop - last_wav_range.start)
+    if coverage < min_pad_coverage and len(mel_slices) > 1:
+        mel_slices = mel_slices[:-1]
+        wav_slices = wav_slices[:-1]
+
+    return wav_slices, mel_slices
+
+
+class SpeakerVerificationPreprocessor(object):
+    def __init__(self,
+                 sampling_rate: int,
+                 audio_norm_target_dBFS: float,
+                 vad_window_length,
+                 vad_moving_average_width,
+                 vad_max_silence_length,
+                 mel_window_length,
+                 mel_window_step,
+                 n_mels,
+                 partial_n_frames: int,
+                 min_pad_coverage: float = 0.75,
+                 partial_overlap_ratio: float = 0.5):
+        self.sampling_rate = sampling_rate
+        self.audio_norm_target_dBFS = audio_norm_target_dBFS
+
+        self.vad_window_length = vad_window_length
+        self.vad_moving_average_width = vad_moving_average_width
+        self.vad_max_silence_length = vad_max_silence_length
+
+        self.n_fft = int(mel_window_length * sampling_rate / 1000)
+        self.hop_length = int(mel_window_step * sampling_rate / 1000)
+        self.n_mels = n_mels
+
+        self.partial_n_frames = partial_n_frames
+        self.min_pad_coverage = min_pad_coverage
+        self.partial_overlap_ratio = partial_overlap_ratio
+
+    def preprocess_wav(self, fpath_or_wav, source_sr=None):
+        # Load the wav from disk if needed
+        if isinstance(fpath_or_wav, (str, Path)):
+            wav, source_sr = librosa.load(str(fpath_or_wav), sr=None)
+        else:
+            wav = fpath_or_wav
+
+        # Resample if numpy.array is passed and sr does not match
+        if source_sr is not None and source_sr != self.sampling_rate:
+            wav = librosa.resample(wav, source_sr, self.sampling_rate)
+
+        # loudness normalization
+        wav = normalize_volume(wav, self.audio_norm_target_dBFS, increase_only=True)
+
+        # trim long silence
+        if webrtcvad:
+            wav = trim_long_silences(wav, self.vad_window_length, self.vad_moving_average_width,
+                                     self.vad_max_silence_length, self.sampling_rate)
+        return wav
+
+    def melspectrogram(self, wav):
+        mel = librosa.feature.melspectrogram(
+            wav, sr=self.sampling_rate, n_fft=self.n_fft, hop_length=self.hop_length, n_mels=self.n_mels)
+        mel = mel.astype(np.float32).T
+        return mel
+
+    def extract_mel_partials(self, wav):
+        wav_slices, mel_slices = compute_partial_slices(
+            len(wav), self.partial_n_frames, self.hop_length, self.min_pad_coverage, self.partial_overlap_ratio)
+
+        # pad audio if needed
+        max_wave_length = wav_slices[-1].stop
+        if max_wave_length >= len(wav):
+            wav = np.pad(wav, (0, max_wave_length - len(wav)), "constant")
+
+        # Split the utterance into partials
+        frames = self.melspectrogram(wav)
+        frames_batch = np.array([frames[s] for s in mel_slices])
+        return frames_batch  # [B, T, C]

modules/text/text_generation/ernie_gen/propeller/paddle/train/__init__.py → modules/audio/voice_cloning/lstm_tacotron2/chinese_g2p.py

-#   Copyright (c) 2019 PaddlePaddle Authors. All Rights Reserved.
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -11,23 +11,29 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
-"""Propeller training"""
 
-from __future__ import print_function
-from __future__ import absolute_import
-from __future__ import unicode_literals
+from typing import List, Tuple
+from pypinyin import lazy_pinyin, Style
 
-import os
-import sys
-import logging
-from time import time
+from .preprocess_transcription import split_syllable
 
-log = logging.getLogger(__name__)
 
-from ernie_gen.propeller.paddle.train.monitored_executor import *
-from ernie_gen.propeller.paddle.train.trainer import *
-from ernie_gen.propeller.paddle.train.hooks import *
-from ernie_gen.propeller.train.model import Model
-from ernie_gen.propeller.paddle.train import exporter
-from ernie_gen.propeller.paddle.train import distribution
-from ernie_gen.propeller.paddle.train import metrics
+def convert_to_pinyin(text: str) -> List[str]:
+    """convert text into list of syllables, other characters that are not chinese, thus
+    cannot be converted to pinyin are splited.
+    """
+    syllables = lazy_pinyin(text, style=Style.TONE3, neutral_tone_with_five=True)
+    return syllables
+
+
+def convert_sentence(text: str) -> List[Tuple[str]]:
+    """convert a sentence into two list: phones and tones"""
+    syllables = convert_to_pinyin(text)
+    phones = []
+    tones = []
+    for syllable in syllables:
+        p, t = split_syllable(syllable)
+        phones.extend(p)
+        tones.extend(t)
+
+    return phones, tones

modules/audio/voice_cloning/lstm_tacotron2/module.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import importlib
+import os
+from typing import List
+
+import numpy as np
+import paddle
+import paddle.nn as nn
+from paddlehub.env import MODULE_HOME
+from paddlehub.module.module import moduleinfo
+from paddlehub.utils.log import logger
+from paddlenlp.data import Pad
+from parakeet.models import ConditionalWaveFlow, Tacotron2
+from parakeet.models.lstm_speaker_encoder import LSTMSpeakerEncoder
+import soundfile as sf
+
+from .audio_processor import SpeakerVerificationPreprocessor
+from .chinese_g2p import convert_sentence
+from .preprocess_transcription import voc_phones, voc_tones, phone_pad_token, tone_pad_token
+
+
+@moduleinfo(
+    name="lstm_tacotron2",
+    version="1.0.0",
+    summary="",
+    author="paddlepaddle",
+    author_email="",
+    type="audio/voice_cloning",
+)
+class VoiceCloner(nn.Layer):
+    def __init__(self, speaker_audio: str = None, output_dir: str = './'):
+        super(VoiceCloner, self).__init__()
+
+        self.sample_rate = 22050  # Hyper params for the following model ckpts.
+        speaker_encoder_ckpt = os.path.join(MODULE_HOME, 'lstm_tacotron2', 'assets',
+                                            'ge2e_ckpt_0.3/step-3000000.pdparams')
+        synthesizer_ckpt = os.path.join(MODULE_HOME, 'lstm_tacotron2', 'assets',
+                                        'tacotron2_aishell3_ckpt_0.3/step-450000.pdparams')
+        vocoder_ckpt = os.path.join(MODULE_HOME, 'lstm_tacotron2', 'assets',
+                                    'waveflow_ljspeech_ckpt_0.3/step-2000000.pdparams')
+
+        # Speaker encoder
+        self.speaker_processor = SpeakerVerificationPreprocessor(
+            sampling_rate=16000,
+            audio_norm_target_dBFS=-30,
+            vad_window_length=30,
+            vad_moving_average_width=8,
+            vad_max_silence_length=6,
+            mel_window_length=25,
+            mel_window_step=10,
+            n_mels=40,
+            partial_n_frames=160,
+            min_pad_coverage=0.75,
+            partial_overlap_ratio=0.5)
+        self.speaker_encoder = LSTMSpeakerEncoder(n_mels=40, num_layers=3, hidden_size=256, output_size=256)
+        self.speaker_encoder.set_state_dict(paddle.load(speaker_encoder_ckpt))
+        self.speaker_encoder.eval()
+
+        # Voice synthesizer
+        self.synthesizer = Tacotron2(
+            vocab_size=68,
+            n_tones=10,
+            d_mels=80,
+            d_encoder=512,
+            encoder_conv_layers=3,
+            encoder_kernel_size=5,
+            d_prenet=256,
+            d_attention_rnn=1024,
+            d_decoder_rnn=1024,
+            attention_filters=32,
+            attention_kernel_size=31,
+            d_attention=128,
+            d_postnet=512,
+            postnet_kernel_size=5,
+            postnet_conv_layers=5,
+            reduction_factor=1,
+            p_encoder_dropout=0.5,
+            p_prenet_dropout=0.5,
+            p_attention_dropout=0.1,
+            p_decoder_dropout=0.1,
+            p_postnet_dropout=0.5,
+            d_global_condition=256,
+            use_stop_token=False)
+        self.synthesizer.set_state_dict(paddle.load(synthesizer_ckpt))
+        self.synthesizer.eval()
+
+        # Vocoder
+        self.vocoder = ConditionalWaveFlow(
+            upsample_factors=[16, 16], n_flows=8, n_layers=8, n_group=16, channels=128, n_mels=80, kernel_size=[3, 3])
+        self.vocoder.set_state_dict(paddle.load(vocoder_ckpt))
+        self.vocoder.eval()
+
+        # Speaking embedding
+        self._speaker_embedding = None
+        if speaker_audio is None or not os.path.isfile(speaker_audio):
+            speaker_audio = os.path.join(MODULE_HOME, 'lstm_tacotron2', 'assets', 'voice_cloning.wav')
+            logger.warning(f'Due to no speaker audio is specified, speaker encoder will use defult '
+                           f'waveform({speaker_audio}) to extract speaker embedding. You can use '
+                           '"set_speaker_embedding()" method to reset a speaker audio for voice cloning.')
+        self.set_speaker_embedding(speaker_audio)
+
+        self.output_dir = os.path.abspath(output_dir)
+        if not os.path.exists(self.output_dir):
+            os.makedirs(self.output_dir)
+
+    def get_speaker_embedding(self):
+        return self._speaker_embedding.numpy()
+
+    def set_speaker_embedding(self, speaker_audio: str):
+        assert os.path.exists(speaker_audio), f'Speaker audio file: {speaker_audio} does not exists.'
+        mel_sequences = self.speaker_processor.extract_mel_partials(
+            self.speaker_processor.preprocess_wav(speaker_audio))
+        self._speaker_embedding = self.speaker_encoder.embed_utterance(paddle.to_tensor(mel_sequences))
+        logger.info(f'Speaker embedding has been set from file: {speaker_audio}')
+
+    def forward(self, phones: paddle.Tensor, tones: paddle.Tensor, speaker_embeddings: paddle.Tensor):
+        outputs = self.synthesizer.infer(phones, tones=tones, global_condition=speaker_embeddings)
+        mel_input = paddle.transpose(outputs["mel_outputs_postnet"], [0, 2, 1])
+        waveforms = self.vocoder.infer(mel_input)
+        return waveforms
+
+    def _convert_text_to_input(self, text: str):
+        """
+        Convert input string to phones and tones.
+        """
+        phones, tones = convert_sentence(text)
+        phones = np.array([voc_phones.lookup(item) for item in phones], dtype=np.int64)
+        tones = np.array([voc_tones.lookup(item) for item in tones], dtype=np.int64)
+        return phones, tones
+
+    def _batchify(self, data: List[str], batch_size: int):
+        """
+        Generate input batches.
+        """
+        phone_pad_func = Pad(voc_phones.lookup(phone_pad_token))
+        tone_pad_func = Pad(voc_tones.lookup(tone_pad_token))
+
+        def _parse_batch(batch_data):
+            phones, tones = zip(*batch_data)
+            speaker_embeddings = paddle.expand(self._speaker_embedding, shape=(len(batch_data), -1))
+            return phone_pad_func(phones), tone_pad_func(tones), speaker_embeddings
+
+        examples = []  # [(phones, tones), ...]
+        for text in data:
+            examples.append(self._convert_text_to_input(text))
+
+        # Seperates data into some batches.
+        one_batch = []
+        for example in examples:
+            one_batch.append(example)
+            if len(one_batch) == batch_size:
+                yield _parse_batch(one_batch)
+                one_batch = []
+        if one_batch:
+            yield _parse_batch(one_batch)
+
+    def generate(self, data: List[str], batch_size: int = 1, use_gpu: bool = False):
+        assert self._speaker_embedding is not None, f'Set speaker embedding before voice cloning.'
+
+        paddle.set_device('gpu') if use_gpu else paddle.set_device('cpu')
+        batches = self._batchify(data, batch_size)
+
+        results = []
+        for batch in batches:
+            phones, tones, speaker_embeddings = map(paddle.to_tensor, batch)
+            waveforms = self(phones, tones, speaker_embeddings).numpy()
+            results.extend(list(waveforms))
+
+        files = []
+        for idx, waveform in enumerate(results):
+            output_wav = os.path.join(self.output_dir, f'{idx+1}.wav')
+            sf.write(output_wav, waveform, samplerate=self.sample_rate)
+            files.append(output_wav)
+
+        return files

modules/audio/voice_cloning/lstm_tacotron2/preprocess_transcription.py

+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import argparse
+from pathlib import Path
+import pickle
+import re
+
+from parakeet.frontend import Vocab
+import tqdm
+
+zh_pattern = re.compile("[\u4e00-\u9fa5]")
+
+_tones = {'<pad>', '<s>', '</s>', '0', '1', '2', '3', '4', '5'}
+
+_pauses = {'%', '$'}
+
+_initials = {
+    'b',
+    'p',
+    'm',
+    'f',
+    'd',
+    't',
+    'n',
+    'l',
+    'g',
+    'k',
+    'h',
+    'j',
+    'q',
+    'x',
+    'zh',
+    'ch',
+    'sh',
+    'r',
+    'z',
+    'c',
+    's',
+}
+
+_finals = {
+    'ii',
+    'iii',
+    'a',
+    'o',
+    'e',
+    'ea',
+    'ai',
+    'ei',
+    'ao',
+    'ou',
+    'an',
+    'en',
+    'ang',
+    'eng',
+    'er',
+    'i',
+    'ia',
+    'io',
+    'ie',
+    'iai',
+    'iao',
+    'iou',
+    'ian',
+    'ien',
+    'iang',
+    'ieng',
+    'u',
+    'ua',
+    'uo',
+    'uai',
+    'uei',
+    'uan',
+    'uen',
+    'uang',
+    'ueng',
+    'v',
+    've',
+    'van',
+    'ven',
+    'veng',
+}
+
+_ernized_symbol = {'&r'}
+
+_specials = {'<pad>', '<unk>', '<s>', '</s>'}
+
+_phones = _initials | _finals | _ernized_symbol | _specials | _pauses
+
+phone_pad_token = '<pad>'
+tone_pad_token = '<pad>'
+voc_phones = Vocab(sorted(list(_phones)))
+voc_tones = Vocab(sorted(list(_tones)))
+
+
+def is_zh(word):
+    global zh_pattern
+    match = zh_pattern.search(word)
+    return match is not None
+
+
+def ernized(syllable):
+    return syllable[:2] != "er" and syllable[-2] == 'r'
+
+
+def convert(syllable):
+    # expansion of o -> uo
+    syllable = re.sub(r"([bpmf])o$", r"\1uo", syllable)
+    # syllable = syllable.replace("bo", "buo").replace("po", "puo").replace("mo", "muo").replace("fo", "fuo")
+    # expansion for iong, ong
+    syllable = syllable.replace("iong", "veng").replace("ong", "ueng")
+
+    # expansion for ing, in
+    syllable = syllable.replace("ing", "ieng").replace("in", "ien")
+
+    # expansion for un, ui, iu
+    syllable = syllable.replace("un", "uen").replace("ui", "uei").replace("iu", "iou")
+
+    # rule for variants of i
+    syllable = syllable.replace("zi", "zii").replace("ci", "cii").replace("si", "sii")\
+        .replace("zhi", "zhiii").replace("chi", "chiii").replace("shi", "shiii")\
+        .replace("ri", "riii")
+
+    # rule for y preceding i, u
+    syllable = syllable.replace("yi", "i").replace("yu", "v").replace("y", "i")
+
+    # rule for w
+    syllable = syllable.replace("wu", "u").replace("w", "u")
+
+    # rule for v following j, q, x
+    syllable = syllable.replace("ju", "jv").replace("qu", "qv").replace("xu", "xv")
+
+    return syllable
+
+
+def split_syllable(syllable: str):
+    """Split a syllable in pinyin into a list of phones and a list of tones.
+    Initials have no tone, represented by '0', while finals have tones from
+    '1,2,3,4,5'.
+
+    e.g.
+
+    zhang -> ['zh', 'ang'], ['0', '1']
+    """
+    if syllable in _pauses:
+        # syllable, tone
+        return [syllable], ['0']
+
+    tone = syllable[-1]
+    syllable = convert(syllable[:-1])
+
+    phones = []
+    tones = []
+
+    global _initials
+    if syllable[:2] in _initials:
+        phones.append(syllable[:2])
+        tones.append('0')
+        phones.append(syllable[2:])
+        tones.append(tone)
+    elif syllable[0] in _initials:
+        phones.append(syllable[0])
+        tones.append('0')
+        phones.append(syllable[1:])
+        tones.append(tone)
+    else:
+        phones.append(syllable)
+        tones.append(tone)
+    return phones, tones

modules/audio/voice_cloning/lstm_tacotron2/requirements.txt

+paddle-parakeet

modules/image/classification/ghostnet_x0_5_imagenet/README.md

+```shell
+$ hub install ghostnet_x0_5_imagenet==1.0.0
+```
+
+## 命令行预测
+
+```shell
+$ hub run ghostnet_x0_5_imagenet --input_path "/PATH/TO/IMAGE" --top_k 5
+```
+
+## 脚本预测
+
+```python
+import paddle
+import paddlehub as hub
+
+if __name__ == '__main__':
+
+    model = hub.Module(name='ghostnet_x0_5_imagenet',)
+    result = model.predict([PATH/TO/IMAGE])
+```
+
+## Fine-tune代码步骤
+
+使用PaddleHub Fine-tune API进行Fine-tune可以分为4个步骤。
+
+### Step1: 定义数据预处理方式
+```python
+import paddlehub.vision.transforms as T
+
+transforms = T.Compose([T.Resize((256, 256)),
+                        T.CenterCrop(224),
+                        T.Normalize(mean=[0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225])],
+                        to_rgb=True)
+```
+
+'transforms' 数据增强模块定义了丰富的数据预处理方式，用户可按照需求替换自己需要的数据预处理方式。
+
+### Step2: 下载数据集并使用
+```python
+from paddlehub.datasets import Flowers
+
+flowers = Flowers(transforms)
+flowers_validate = Flowers(transforms, mode='val')
+```
+* transforms(Callable): 数据预处理方式。
+* mode(str): 选择数据模式，可选项有 'train', 'test', 'val'， 默认为'train'。
+
+'hub.datasets.Flowers()' 会自动从网络下载数据集并解压到用户目录下'$HOME/.paddlehub/dataset'目录。
+
+
+### Step3: 加载预训练模型
+
+```python
+import paddlehub as hub
+
+model = hub.Module(name='ghostnet_x0_5_imagenet',
+                   label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"],
+                   load_checkpoint=None)
+```
+* name(str): 选择预训练模型的名字。
+* label_list(list): 设置标签对应分类类别, 默认为Imagenet2012类别。
+* load _checkpoint(str): 模型参数地址。
+
+PaddleHub提供许多图像分类预训练模型，如xception、mobilenet、efficientnet等，详细信息参见[图像分类模型](https://www.paddlepaddle.org.cn/hub?filter=en_category&value=ImageClassification)。
+
+如果想尝试efficientnet模型，只需要更换Module中的'name'参数即可.
+```python
+import paddlehub as hub
+
+# 更换name参数即可无缝切换efficientnet模型, 代码示例如下
+module = hub.Module(name="efficientnetb7_imagenet")
+```
+**NOTE:**目前部分模型还没有完全升级到2.0版本，敬请期待。
+
+### Step4: 选择优化策略和运行配置
+
+```python
+import paddle
+from paddlehub.finetune.trainer import Trainer
+
+optimizer = paddle.optimizer.Adam(learning_rate=0.001, parameters=model.parameters())
+trainer = Trainer(model, optimizer, checkpoint_dir='img_classification_ckpt')
+
+trainer.train(flowers, epochs=100, batch_size=32, eval_dataset=flowers_validate, save_interval=1)
+```
+
+#### 优化策略
+
+Paddle2.0rc提供了多种优化器选择，如'SGD', 'Adam', 'Adamax'等，详细参见[策略](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/optimizer/optimizer/Optimizer_cn.html)。
+
+其中'Adam':
+
+* learning_rate: 全局学习率。默认为1e-3；
+* parameters: 待优化模型参数。
+
+#### 运行配置
+'Trainer' 主要控制Fine-tune的训练，包含以下可控制的参数:
+
+* model: 被优化模型；
+* optimizer: 优化器选择；
+* use_vdl: 是否使用vdl可视化训练过程；
+* checkpoint_dir: 保存模型参数的地址；
+* compare_metrics: 保存最优模型的衡量指标；
+
+'trainer.train' 主要控制具体的训练过程，包含以下可控制的参数：
+
+* train_dataset: 训练时所用的数据集；
+* epochs: 训练轮数；
+* batch_size: 训练的批大小，如果使用GPU，请根据实际情况调整batch_size；
+* num_workers: works的数量，默认为0；
+* eval_dataset: 验证集；
+* log_interval: 打印日志的间隔， 单位为执行批训练的次数。
+* save_interval: 保存模型的间隔频次，单位为执行训练的轮数。
+
+## 模型预测
+
+当完成Fine-tune后，Fine-tune过程在验证集上表现最优的模型会被保存在'${CHECKPOINT_DIR}/best_model'目录下，其中'${CHECKPOINT_DIR}'目录为Fine-tune时所选择的保存checkpoint的目录。
+我们使用该模型来进行预测。predict.py脚本如下：
+
+```python
+import paddle
+import paddlehub as hub
+
+if __name__ == '__main__':
+
+    model = hub.Module(name='ghostnet_x0_5_imagenet', label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"], load_checkpoint='/PATH/TO/CHECKPOINT')
+    result = model.predict(['flower.jpg'])
+```
+
+参数配置正确后，请执行脚本'python predict.py'， 加载模型具体可参见[加载](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/framework/io/load_cn.html#load)。
+
+**NOTE:** 进行预测时，所选择的module，checkpoint_dir，dataset必须和Fine-tune所用的一样。
+
+## 服务部署
+
+PaddleHub Serving可以部署一个在线分类任务服务
+
+## Step1: 启动PaddleHub Serving
+
+运行启动命令：
+
+```shell
+$ hub serving start -m ghostnet_x0_5_imagenet
+```
+
+这样就完成了一个分类任务服务化API的部署，默认端口号为8866。
+
+**NOTE:** 如使用GPU预测，则需要在启动服务之前，请设置CUDA_VISIBLE_DEVICES环境变量，否则不用设置。
+
+## Step2: 发送预测请求
+
+配置好服务端，以下数行代码即可实现发送预测请求，获取预测结果
+
+```python
+import requests
+import json
+import cv2
+import base64
+
+import numpy as np
+
+
+def cv2_to_base64(image):
+    data = cv2.imencode('.jpg', image)[1]
+    return base64.b64encode(data.tostring()).decode('utf8')
+
+def base64_to_cv2(b64str):
+    data = base64.b64decode(b64str.encode('utf8'))
+    data = np.fromstring(data, np.uint8)
+    data = cv2.imdecode(data, cv2.IMREAD_COLOR)
+    return data
+
+# 发送HTTP请求
+org_im = cv2.imread('/PATH/TO/IMAGE')
+
+data = {'images':[cv2_to_base64(org_im)], 'top_k':2}
+headers = {"Content-type": "application/json"}
+url = "http://127.0.0.1:8866/predict/ghostnet_x0_5_imagenet"
+r = requests.post(url=url, headers=headers, data=json.dumps(data))
+data =r.json()["results"]['data']
+```
+
+### 查看代码
+
+[PaddleClas](https://github.com/PaddlePaddle/PaddleClas)
+
+### 依赖
+
+paddlepaddle >= 2.0.0
+
+paddlehub >= 2.0.0

modules/image/classification/ghostnet_x0_5_imagenet/label_list.txt

+tench
+goldfish
+great white shark
+tiger shark
+hammerhead
+electric ray
+stingray
+cock
+hen
+ostrich
+brambling
+goldfinch
+house finch
+junco
+indigo bunting
+robin
+bulbul
+jay
+magpie
+chickadee
+water ouzel
+kite
+bald eagle
+vulture
+great grey owl
+European fire salamander
+common newt
+eft
+spotted salamander
+axolotl
+bullfrog
+tree frog
+tailed frog
+loggerhead
+leatherback turtle
+mud turtle
+terrapin
+box turtle
+banded gecko
+common iguana
+American chameleon
+whiptail
+agama
+frilled lizard
+alligator lizard
+Gila monster
+green lizard
+African chameleon
+Komodo dragon
+African crocodile
+American alligator
+triceratops
+thunder snake
+ringneck snake
+hognose snake
+green snake
+king snake
+garter snake
+water snake
+vine snake
+night snake
+boa constrictor
+rock python
+Indian cobra
+green mamba
+sea snake
+horned viper
+diamondback
+sidewinder
+trilobite
+harvestman
+scorpion
+black and gold garden spider
+barn spider
+garden spider
+black widow
+tarantula
+wolf spider
+tick
+centipede
+black grouse
+ptarmigan
+ruffed grouse
+prairie chicken
+peacock
+quail
+partridge
+African grey
+macaw
+sulphur-crested cockatoo
+lorikeet
+coucal
+bee eater
+hornbill
+hummingbird
+jacamar
+toucan
+drake
+red-breasted merganser
+goose
+black swan
+tusker
+echidna
+platypus
+wallaby
+koala
+wombat
+jellyfish
+sea anemone
+brain coral
+flatworm
+nematode
+conch
+snail
+slug
+sea slug
+chiton
+chambered nautilus
+Dungeness crab
+rock crab
+fiddler crab
+king crab
+American lobster
+spiny lobster
+crayfish
+hermit crab
+isopod
+white stork
+black stork
+spoonbill
+flamingo
+little blue heron
+American egret
+bittern
+crane
+limpkin
+European gallinule
+American coot
+bustard
+ruddy turnstone
+red-backed sandpiper
+redshank
+dowitcher
+oystercatcher
+pelican
+king penguin
+albatross
+grey whale
+killer whale
+dugong
+sea lion
+Chihuahua
+Japanese spaniel
+Maltese dog
+Pekinese
+Shih-Tzu
+Blenheim spaniel
+papillon
+toy terrier
+Rhodesian ridgeback
+Afghan hound
+basset
+beagle
+bloodhound
+bluetick
+black-and-tan coonhound
+Walker hound
+English foxhound
+redbone
+borzoi
+Irish wolfhound
+Italian greyhound
+whippet
+Ibizan hound
+Norwegian elkhound
+otterhound
+Saluki
+Scottish deerhound
+Weimaraner
+Staffordshire bullterrier
+American Staffordshire terrier
+Bedlington terrier
+Border terrier
+Kerry blue terrier
+Irish terrier
+Norfolk terrier
+Norwich terrier
+Yorkshire terrier
+wire-haired fox terrier
+Lakeland terrier
+Sealyham terrier
+Airedale
+cairn
+Australian terrier
+Dandie Dinmont
+Boston bull
+miniature schnauzer
+giant schnauzer
+standard schnauzer
+Scotch terrier
+Tibetan terrier
+silky terrier
+soft-coated wheaten terrier
+West Highland white terrier
+Lhasa
+flat-coated retriever
+curly-coated retriever
+golden retriever
+Labrador retriever
+Chesapeake Bay retriever
+German short-haired pointer
+vizsla
+English setter
+Irish setter
+Gordon setter
+Brittany spaniel
+clumber
+English springer
+Welsh springer spaniel
+cocker spaniel
+Sussex spaniel
+Irish water spaniel
+kuvasz
+schipperke
+groenendael
+malinois
+briard
+kelpie
+komondor
+Old English sheepdog
+Shetland sheepdog
+collie
+Border collie
+Bouvier des Flandres
+Rottweiler
+German shepherd
+Doberman
+miniature pinscher
+Greater Swiss Mountain dog
+Bernese mountain dog
+Appenzeller
+EntleBucher
+boxer
+bull mastiff
+Tibetan mastiff
+French bulldog
+Great Dane
+Saint Bernard
+Eskimo dog
+malamute
+Siberian husky
+dalmatian
+affenpinscher
+basenji
+pug
+Leonberg
+Newfoundland
+Great Pyrenees
+Samoyed
+Pomeranian
+chow
+keeshond
+Brabancon griffon
+Pembroke
+Cardigan
+toy poodle
+miniature poodle
+standard poodle
+Mexican hairless
+timber wolf
+white wolf
+red wolf
+coyote
+dingo
+dhole
+African hunting dog
+hyena
+red fox
+kit fox
+Arctic fox
+grey fox
+tabby
+tiger cat
+Persian cat
+Siamese cat
+Egyptian cat
+cougar
+lynx
+leopard
+snow leopard
+jaguar
+lion
+tiger
+cheetah
+brown bear
+American black bear
+ice bear
+sloth bear
+mongoose
+meerkat
+tiger beetle
+ladybug
+ground beetle
+long-horned beetle
+leaf beetle
+dung beetle
+rhinoceros beetle
+weevil
+fly
+bee
+ant
+grasshopper
+cricket
+walking stick
+cockroach
+mantis
+cicada
+leafhopper
+lacewing
+dragonfly
+damselfly
+admiral
+ringlet
+monarch
+cabbage butterfly
+sulphur butterfly
+lycaenid
+starfish
+sea urchin
+sea cucumber
+wood rabbit
+hare
+Angora
+hamster
+porcupine
+fox squirrel
+marmot
+beaver
+guinea pig
+sorrel
+zebra
+hog
+wild boar
+warthog
+hippopotamus
+ox
+water buffalo
+bison
+ram
+bighorn
+ibex
+hartebeest
+impala
+gazelle
+Arabian camel
+llama
+weasel
+mink
+polecat
+black-footed ferret
+otter
+skunk
+badger
+armadillo
+three-toed sloth
+orangutan
+gorilla
+chimpanzee
+gibbon
+siamang
+guenon
+patas
+baboon
+macaque
+langur
+colobus
+proboscis monkey
+marmoset
+capuchin
+howler monkey
+titi
+spider monkey
+squirrel monkey
+Madagascar cat
+indri
+Indian elephant
+African elephant
+lesser panda
+giant panda
+barracouta
+eel
+coho
+rock beauty
+anemone fish
+sturgeon
+gar
+lionfish
+puffer
+abacus
+abaya
+academic gown
+accordion
+acoustic guitar
+aircraft carrier
+airliner
+airship
+altar
+ambulance
+amphibian
+analog clock
+apiary
+apron
+ashcan
+assault rifle
+backpack
+bakery
+balance beam
+balloon
+ballpoint
+Band Aid
+banjo
+bannister
+barbell
+barber chair
+barbershop
+barn
+barometer
+barrel
+barrow
+baseball
+basketball
+bassinet
+bassoon
+bathing cap
+bath towel
+bathtub
+beach wagon
+beacon
+beaker
+bearskin
+beer bottle
+beer glass
+bell cote
+bib
+bicycle-built-for-two
+bikini
+binder
+binoculars
+birdhouse
+boathouse
+bobsled
+bolo tie
+bonnet
+bookcase
+bookshop
+bottlecap
+bow
+bow tie
+brass
+brassiere
+breakwater
+breastplate
+broom
+bucket
+buckle
+bulletproof vest
+bullet train
+butcher shop
+cab
+caldron
+candle
+cannon
+canoe
+can opener
+cardigan
+car mirror
+carousel
+carpenters kit
+carton
+car wheel
+cash machine
+cassette
+cassette player
+castle
+catamaran
+CD player
+cello
+cellular telephone
+chain
+chainlink fence
+chain mail
+chain saw
+chest
+chiffonier
+chime
+china cabinet
+Christmas stocking
+church
+cinema
+cleaver
+cliff dwelling
+cloak
+clog
+cocktail shaker
+coffee mug
+coffeepot
+coil
+combination lock
+computer keyboard
+confectionery
+container ship
+convertible
+corkscrew
+cornet
+cowboy boot
+cowboy hat
+cradle
+crane
+crash helmet
+crate
+crib
+Crock Pot
+croquet ball
+crutch
+cuirass
+dam
+desk
+desktop computer
+dial telephone
+diaper
+digital clock
+digital watch
+dining table
+dishrag
+dishwasher
+disk brake
+dock
+dogsled
+dome
+doormat
+drilling platform
+drum
+drumstick
+dumbbell
+Dutch oven
+electric fan
+electric guitar
+electric locomotive
+entertainment center
+envelope
+espresso maker
+face powder
+feather boa
+file
+fireboat
+fire engine
+fire screen
+flagpole
+flute
+folding chair
+football helmet
+forklift
+fountain
+fountain pen
+four-poster
+freight car
+French horn
+frying pan
+fur coat
+garbage truck
+gasmask
+gas pump
+goblet
+go-kart
+golf ball
+golfcart
+gondola
+gong
+gown
+grand piano
+greenhouse
+grille
+grocery store
+guillotine
+hair slide
+hair spray
+half track
+hammer
+hamper
+hand blower
+hand-held computer
+handkerchief
+hard disc
+harmonica
+harp
+harvester
+hatchet
+holster
+home theater
+honeycomb
+hook
+hoopskirt
+horizontal bar
+horse cart
+hourglass
+iPod
+iron
+jack-o-lantern
+jean
+jeep
+jersey
+jigsaw puzzle
+jinrikisha
+joystick
+kimono
+knee pad
+knot
+lab coat
+ladle
+lampshade
+laptop
+lawn mower
+lens cap
+letter opener
+library
+lifeboat
+lighter
+limousine
+liner
+lipstick
+Loafer
+lotion
+loudspeaker
+loupe
+lumbermill
+magnetic compass
+mailbag
+mailbox
+maillot
+maillot
+manhole cover
+maraca
+marimba
+mask
+matchstick
+maypole
+maze
+measuring cup
+medicine chest
+megalith
+microphone
+microwave
+military uniform
+milk can
+minibus
+miniskirt
+minivan
+missile
+mitten
+mixing bowl
+mobile home
+Model T
+modem
+monastery
+monitor
+moped
+mortar
+mortarboard
+mosque
+mosquito net
+motor scooter
+mountain bike
+mountain tent
+mouse
+mousetrap
+moving van
+muzzle
+nail
+neck brace
+necklace
+nipple
+notebook
+obelisk
+oboe
+ocarina
+odometer
+oil filter
+organ
+oscilloscope
+overskirt
+oxcart
+oxygen mask
+packet
+paddle
+paddlewheel
+padlock
+paintbrush
+pajama
+palace
+panpipe
+paper towel
+parachute
+parallel bars
+park bench
+parking meter
+passenger car
+patio
+pay-phone
+pedestal
+pencil box
+pencil sharpener
+perfume
+Petri dish
+photocopier
+pick
+pickelhaube
+picket fence
+pickup
+pier
+piggy bank
+pill bottle
+pillow
+ping-pong ball
+pinwheel
+pirate
+pitcher
+plane
+planetarium
+plastic bag
+plate rack
+plow
+plunger
+Polaroid camera
+pole
+police van
+poncho
+pool table
+pop bottle
+pot
+potters wheel
+power drill
+prayer rug
+printer
+prison
+projectile
+projector
+puck
+punching bag
+purse
+quill
+quilt
+racer
+racket
+radiator
+radio
+radio telescope
+rain barrel
+recreational vehicle
+reel
+reflex camera
+refrigerator
+remote control
+restaurant
+revolver
+rifle
+rocking chair
+rotisserie
+rubber eraser
+rugby ball
+rule
+running shoe
+safe
+safety pin
+saltshaker
+sandal
+sarong
+sax
+scabbard
+scale
+school bus
+schooner
+scoreboard
+screen
+screw
+screwdriver
+seat belt
+sewing machine
+shield
+shoe shop
+shoji
+shopping basket
+shopping cart
+shovel
+shower cap
+shower curtain
+ski
+ski mask
+sleeping bag
+slide rule
+sliding door
+slot
+snorkel
+snowmobile
+snowplow
+soap dispenser
+soccer ball
+sock
+solar dish
+sombrero
+soup bowl
+space bar
+space heater
+space shuttle
+spatula
+speedboat
+spider web
+spindle
+sports car
+spotlight
+stage
+steam locomotive
+steel arch bridge
+steel drum
+stethoscope
+stole
+stone wall
+stopwatch
+stove
+strainer
+streetcar
+stretcher
+studio couch
+stupa
+submarine
+suit
+sundial
+sunglass
+sunglasses
+sunscreen
+suspension bridge
+swab
+sweatshirt
+swimming trunks
+swing
+switch
+syringe
+table lamp
+tank
+tape player
+teapot
+teddy
+television
+tennis ball
+thatch
+theater curtain
+thimble
+thresher
+throne
+tile roof
+toaster
+tobacco shop
+toilet seat
+torch
+totem pole
+tow truck
+toyshop
+tractor
+trailer truck
+tray
+trench coat
+tricycle
+trimaran
+tripod
+triumphal arch
+trolleybus
+trombone
+tub
+turnstile
+typewriter keyboard
+umbrella
+unicycle
+upright
+vacuum
+vase
+vault
+velvet
+vending machine
+vestment
+viaduct
+violin
+volleyball
+waffle iron
+wall clock
+wallet
+wardrobe
+warplane
+washbasin
+washer
+water bottle
+water jug
+water tower
+whiskey jug
+whistle
+wig
+window screen
+window shade
+Windsor tie
+wine bottle
+wing
+wok
+wooden spoon
+wool
+worm fence
+wreck
+yawl
+yurt
+web site
+comic book
+crossword puzzle
+street sign
+traffic light
+book jacket
+menu
+plate
+guacamole
+consomme
+hot pot
+trifle
+ice cream
+ice lolly
+French loaf
+bagel
+pretzel
+cheeseburger
+hotdog
+mashed potato
+head cabbage
+broccoli
+cauliflower
+zucchini
+spaghetti squash
+acorn squash
+butternut squash
+cucumber
+artichoke
+bell pepper
+cardoon
+mushroom
+Granny Smith
+strawberry
+orange
+lemon
+fig
+pineapple
+banana
+jackfruit
+custard apple
+pomegranate
+hay
+carbonara
+chocolate sauce
+dough
+meat loaf
+pizza
+potpie
+burrito
+red wine
+espresso
+cup
+eggnog
+alp
+bubble
+cliff
+coral reef
+geyser
+lakeside
+promontory
+sandbar
+seashore
+valley
+volcano
+ballplayer
+groom
+scuba diver
+rapeseed
+daisy
+yellow ladys slipper
+corn
+acorn
+hip
+buckeye
+coral fungus
+agaric
+gyromitra
+stinkhorn
+earthstar
+hen-of-the-woods
+bolete
+ear
+toilet tissue

modules/image/classification/ghostnet_x0_5_imagenet/module.py

+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import math
+from typing import Union
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+import paddlehub.vision.transforms as T
+import numpy as np
+from paddle import ParamAttr
+from paddle.nn.initializer import Uniform, KaimingNormal
+from paddlehub.module.module import moduleinfo
+from paddlehub.module.cv_module import ImageClassifierModule
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(self, in_channels, out_channels, kernel_size, stride=1, groups=1, act="relu", name=None):
+        super(ConvBNLayer, self).__init__()
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal(), name=name + "_weights"),
+            bias_attr=False)
+        bn_name = name + "_bn"
+
+        self._batch_norm = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale", regularizer=paddle.regularizer.L2Decay(0.0)),
+            bias_attr=ParamAttr(name=bn_name + "_offset", regularizer=paddle.regularizer.L2Decay(0.0)),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance")
+
+    def forward(self, inputs):
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class SEBlock(nn.Layer):
+    def __init__(self, num_channels, reduction_ratio=4, name=None):
+        super(SEBlock, self).__init__()
+        self.pool2d_gap = nn.AdaptiveAvgPool2D(1)
+        self._num_channels = num_channels
+        stdv = 1.0 / math.sqrt(num_channels * 1.0)
+        med_ch = num_channels // reduction_ratio
+        self.squeeze = nn.Linear(
+            num_channels,
+            med_ch,
+            weight_attr=ParamAttr(initializer=Uniform(-stdv, stdv), name=name + "_1_weights"),
+            bias_attr=ParamAttr(name=name + "_1_offset"))
+        stdv = 1.0 / math.sqrt(med_ch * 1.0)
+        self.excitation = nn.Linear(
+            med_ch,
+            num_channels,
+            weight_attr=ParamAttr(initializer=Uniform(-stdv, stdv), name=name + "_2_weights"),
+            bias_attr=ParamAttr(name=name + "_2_offset"))
+
+    def forward(self, inputs):
+        pool = self.pool2d_gap(inputs)
+        pool = paddle.squeeze(pool, axis=[2, 3])
+        squeeze = self.squeeze(pool)
+        squeeze = F.relu(squeeze)
+        excitation = self.excitation(squeeze)
+        excitation = paddle.clip(x=excitation, min=0, max=1)
+        excitation = paddle.unsqueeze(excitation, axis=[2, 3])
+        out = paddle.multiply(inputs, excitation)
+        return out
+
+
+class GhostModule(nn.Layer):
+    def __init__(self, in_channels, output_channels, kernel_size=1, ratio=2, dw_size=3, stride=1, relu=True, name=None):
+        super(GhostModule, self).__init__()
+        init_channels = int(math.ceil(output_channels / ratio))
+        new_channels = int(init_channels * (ratio - 1))
+        self.primary_conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=init_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            groups=1,
+            act="relu" if relu else None,
+            name=name + "_primary_conv")
+        self.cheap_operation = ConvBNLayer(
+            in_channels=init_channels,
+            out_channels=new_channels,
+            kernel_size=dw_size,
+            stride=1,
+            groups=init_channels,
+            act="relu" if relu else None,
+            name=name + "_cheap_operation")
+
+    def forward(self, inputs):
+        x = self.primary_conv(inputs)
+        y = self.cheap_operation(x)
+        out = paddle.concat([x, y], axis=1)
+        return out
+
+
+class GhostBottleneck(nn.Layer):
+    def __init__(self, in_channels, hidden_dim, output_channels, kernel_size, stride, use_se, name=None):
+        super(GhostBottleneck, self).__init__()
+        self._stride = stride
+        self._use_se = use_se
+        self._num_channels = in_channels
+        self._output_channels = output_channels
+        self.ghost_module_1 = GhostModule(
+            in_channels=in_channels,
+            output_channels=hidden_dim,
+            kernel_size=1,
+            stride=1,
+            relu=True,
+            name=name + "_ghost_module_1")
+        if stride == 2:
+            self.depthwise_conv = ConvBNLayer(
+                in_channels=hidden_dim,
+                out_channels=hidden_dim,
+                kernel_size=kernel_size,
+                stride=stride,
+                groups=hidden_dim,
+                act=None,
+                name=name + "_depthwise_depthwise"  # looks strange due to an old typo, will be fixed later.
+            )
+        if use_se:
+            self.se_block = SEBlock(num_channels=hidden_dim, name=name + "_se")
+        self.ghost_module_2 = GhostModule(
+            in_channels=hidden_dim,
+            output_channels=output_channels,
+            kernel_size=1,
+            relu=False,
+            name=name + "_ghost_module_2")
+        if stride != 1 or in_channels != output_channels:
+            self.shortcut_depthwise = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=in_channels,
+                kernel_size=kernel_size,
+                stride=stride,
+                groups=in_channels,
+                act=None,
+                name=name + "_shortcut_depthwise_depthwise"  # looks strange due to an old typo, will be fixed later.
+            )
+            self.shortcut_conv = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=output_channels,
+                kernel_size=1,
+                stride=1,
+                groups=1,
+                act=None,
+                name=name + "_shortcut_conv")
+
+    def forward(self, inputs):
+        x = self.ghost_module_1(inputs)
+        if self._stride == 2:
+            x = self.depthwise_conv(x)
+        if self._use_se:
+            x = self.se_block(x)
+        x = self.ghost_module_2(x)
+        if self._stride == 1 and self._num_channels == self._output_channels:
+            shortcut = inputs
+        else:
+            shortcut = self.shortcut_depthwise(inputs)
+            shortcut = self.shortcut_conv(shortcut)
+        return paddle.add(x=x, y=shortcut)
+
+
+@moduleinfo(
+    name="ghostnet_x0_5_imagenet",
+    type="CV/classification",
+    author="paddlepaddle",
+    author_email="",
+    summary="ghostnet_x0_5_imagenet is a classification model, "
+    "this module is trained with Imagenet dataset.",
+    version="1.0.0",
+    meta=ImageClassifierModule)
+class GhostNet(nn.Layer):
+    def __init__(self, label_list: list = None, load_checkpoint: str = None):
+        super(GhostNet, self).__init__()
+
+        if label_list is not None:
+            self.labels = label_list
+            class_dim = len(self.labels)
+        else:
+            label_list = []
+            label_file = os.path.join(self.directory, 'label_list.txt')
+            files = open(label_file)
+            for line in files.readlines():
+                line = line.strip('\n')
+                label_list.append(line)
+            self.labels = label_list
+            class_dim = len(self.labels)
+
+        self.cfgs = [
+            # k, t, c, SE, s
+            [3, 16, 16, 0, 1],
+            [3, 48, 24, 0, 2],
+            [3, 72, 24, 0, 1],
+            [5, 72, 40, 1, 2],
+            [5, 120, 40, 1, 1],
+            [3, 240, 80, 0, 2],
+            [3, 200, 80, 0, 1],
+            [3, 184, 80, 0, 1],
+            [3, 184, 80, 0, 1],
+            [3, 480, 112, 1, 1],
+            [3, 672, 112, 1, 1],
+            [5, 672, 160, 1, 2],
+            [5, 960, 160, 0, 1],
+            [5, 960, 160, 1, 1],
+            [5, 960, 160, 0, 1],
+            [5, 960, 160, 1, 1]
+        ]
+        self.scale = 0.5
+        output_channels = int(self._make_divisible(16 * self.scale, 4))
+        self.conv1 = ConvBNLayer(
+            in_channels=3, out_channels=output_channels, kernel_size=3, stride=2, groups=1, act="relu", name="conv1")
+        # build inverted residual blocks
+        idx = 0
+        self.ghost_bottleneck_list = []
+        for k, exp_size, c, use_se, s in self.cfgs:
+            in_channels = output_channels
+            output_channels = int(self._make_divisible(c * self.scale, 4))
+            hidden_dim = int(self._make_divisible(exp_size * self.scale, 4))
+            ghost_bottleneck = self.add_sublayer(
+                name="_ghostbottleneck_" + str(idx),
+                sublayer=GhostBottleneck(
+                    in_channels=in_channels,
+                    hidden_dim=hidden_dim,
+                    output_channels=output_channels,
+                    kernel_size=k,
+                    stride=s,
+                    use_se=use_se,
+                    name="_ghostbottleneck_" + str(idx)))
+            self.ghost_bottleneck_list.append(ghost_bottleneck)
+            idx += 1
+        # build last several layers
+        in_channels = output_channels
+        output_channels = int(self._make_divisible(exp_size * self.scale, 4))
+        self.conv_last = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=output_channels,
+            kernel_size=1,
+            stride=1,
+            groups=1,
+            act="relu",
+            name="conv_last")
+        self.pool2d_gap = nn.AdaptiveAvgPool2D(1)
+        in_channels = output_channels
+        self._fc0_output_channels = 1280
+        self.fc_0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=self._fc0_output_channels,
+            kernel_size=1,
+            stride=1,
+            act="relu",
+            name="fc_0")
+        self.dropout = nn.Dropout(p=0.2)
+        stdv = 1.0 / math.sqrt(self._fc0_output_channels * 1.0)
+        self.fc_1 = nn.Linear(
+            self._fc0_output_channels,
+            class_dim,
+            weight_attr=ParamAttr(name="fc_1_weights", initializer=Uniform(-stdv, stdv)),
+            bias_attr=ParamAttr(name="fc_1_offset"))
+
+        if load_checkpoint is not None:
+            self.model_dict = paddle.load(load_checkpoint)
+            self.set_dict(self.model_dict)
+            print("load custom checkpoint success")
+        else:
+            checkpoint = os.path.join(self.directory, 'model.pdparams')
+            self.model_dict = paddle.load(checkpoint)
+            self.set_dict(self.model_dict)
+            print("load pretrained checkpoint success")
+
+    def transforms(self, images: Union[str, np.ndarray]):
+        transforms = T.Compose([
+            T.Resize((256, 256)),
+            T.CenterCrop(224),
+            T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+        ],
+                               to_rgb=True)
+        return transforms(images).astype('float32')
+
+    def forward(self, inputs):
+        x = self.conv1(inputs)
+        for ghost_bottleneck in self.ghost_bottleneck_list:
+            x = ghost_bottleneck(x)
+        x = self.conv_last(x)
+        feature = self.pool2d_gap(x)
+        x = self.fc_0(feature)
+        x = self.dropout(x)
+        x = paddle.reshape(x, shape=[-1, self._fc0_output_channels])
+        x = self.fc_1(x)
+        return x, feature
+
+    def _make_divisible(self, v, divisor, min_value=None):
+        """
+        This function is taken from the original tf repo.
+        It ensures that all layers have a channel number that is divisible by 8
+        It can be seen here:
+        https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
+        """
+        if min_value is None:
+            min_value = divisor
+        new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+        # Make sure that round down does not go down by more than 10%.
+        if new_v < 0.9 * v:
+            new_v += divisor
+        return new_v

modules/image/classification/ghostnet_x1_0_imagenet/README.md

+```shell
+$ hub install ghostnet_x1_0_imagenet==1.0.0
+```
+
+## 命令行预测
+
+```shell
+$ hub run ghostnet_x1_0_imagenet --input_path "/PATH/TO/IMAGE" --top_k 5
+```
+
+## 脚本预测
+
+```python
+import paddle
+import paddlehub as hub
+
+if __name__ == '__main__':
+
+    model = hub.Module(name='ghostnet_x1_0_imagenet',)
+    result = model.predict([PATH/TO/IMAGE])
+```
+
+## Fine-tune代码步骤
+
+使用PaddleHub Fine-tune API进行Fine-tune可以分为4个步骤。
+
+### Step1: 定义数据预处理方式
+```python
+import paddlehub.vision.transforms as T
+
+transforms = T.Compose([T.Resize((256, 256)),
+                        T.CenterCrop(224),
+                        T.Normalize(mean=[0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225])],
+                        to_rgb=True)
+```
+
+'transforms' 数据增强模块定义了丰富的数据预处理方式，用户可按照需求替换自己需要的数据预处理方式。
+
+### Step2: 下载数据集并使用
+```python
+from paddlehub.datasets import Flowers
+
+flowers = Flowers(transforms)
+flowers_validate = Flowers(transforms, mode='val')
+```
+* transforms(Callable): 数据预处理方式。
+* mode(str): 选择数据模式，可选项有 'train', 'test', 'val'， 默认为'train'。
+
+'hub.datasets.Flowers()' 会自动从网络下载数据集并解压到用户目录下'$HOME/.paddlehub/dataset'目录。
+
+
+### Step3: 加载预训练模型
+
+```python
+import paddlehub as hub
+
+model = hub.Module(name='ghostnet_x1_0_imagenet',
+                   label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"],
+                   load_checkpoint=None)
+```
+* name(str): 选择预训练模型的名字。
+* label_list(list): 设置标签对应分类类别, 默认为Imagenet2012类别。
+* load _checkpoint(str): 模型参数地址。
+
+PaddleHub提供许多图像分类预训练模型，如xception、mobilenet、efficientnet等，详细信息参见[图像分类模型](https://www.paddlepaddle.org.cn/hub?filter=en_category&value=ImageClassification)。
+
+如果想尝试efficientnet模型，只需要更换Module中的'name'参数即可.
+```python
+import paddlehub as hub
+
+# 更换name参数即可无缝切换efficientnet模型, 代码示例如下
+module = hub.Module(name="efficientnetb7_imagenet")
+```
+**NOTE:**目前部分模型还没有完全升级到2.0版本，敬请期待。
+
+### Step4: 选择优化策略和运行配置
+
+```python
+import paddle
+from paddlehub.finetune.trainer import Trainer
+
+optimizer = paddle.optimizer.Adam(learning_rate=0.001, parameters=model.parameters())
+trainer = Trainer(model, optimizer, checkpoint_dir='img_classification_ckpt')
+
+trainer.train(flowers, epochs=100, batch_size=32, eval_dataset=flowers_validate, save_interval=1)
+```
+
+#### 优化策略
+
+Paddle2.0rc提供了多种优化器选择，如'SGD', 'Adam', 'Adamax'等，详细参见[策略](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/optimizer/optimizer/Optimizer_cn.html)。
+
+其中'Adam':
+
+* learning_rate: 全局学习率。默认为1e-3；
+* parameters: 待优化模型参数。
+
+#### 运行配置
+'Trainer' 主要控制Fine-tune的训练，包含以下可控制的参数:
+
+* model: 被优化模型；
+* optimizer: 优化器选择；
+* use_vdl: 是否使用vdl可视化训练过程；
+* checkpoint_dir: 保存模型参数的地址；
+* compare_metrics: 保存最优模型的衡量指标；
+
+'trainer.train' 主要控制具体的训练过程，包含以下可控制的参数：
+
+* train_dataset: 训练时所用的数据集；
+* epochs: 训练轮数；
+* batch_size: 训练的批大小，如果使用GPU，请根据实际情况调整batch_size；
+* num_workers: works的数量，默认为0；
+* eval_dataset: 验证集；
+* log_interval: 打印日志的间隔， 单位为执行批训练的次数。
+* save_interval: 保存模型的间隔频次，单位为执行训练的轮数。
+
+## 模型预测
+
+当完成Fine-tune后，Fine-tune过程在验证集上表现最优的模型会被保存在'${CHECKPOINT_DIR}/best_model'目录下，其中'${CHECKPOINT_DIR}'目录为Fine-tune时所选择的保存checkpoint的目录。
+我们使用该模型来进行预测。predict.py脚本如下：
+
+```python
+import paddle
+import paddlehub as hub
+
+if __name__ == '__main__':
+
+    model = hub.Module(name='ghostnet_x1_0_imagenet', label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"], load_checkpoint='/PATH/TO/CHECKPOINT')
+    result = model.predict(['flower.jpg'])
+```
+
+参数配置正确后，请执行脚本'python predict.py'， 加载模型具体可参见[加载](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/framework/io/load_cn.html#load)。
+
+**NOTE:** 进行预测时，所选择的module，checkpoint_dir，dataset必须和Fine-tune所用的一样。
+
+## 服务部署
+
+PaddleHub Serving可以部署一个在线分类任务服务
+
+## Step1: 启动PaddleHub Serving
+
+运行启动命令：
+
+```shell
+$ hub serving start -m ghostnet_x1_0_imagenet
+```
+
+这样就完成了一个分类任务服务化API的部署，默认端口号为8866。
+
+**NOTE:** 如使用GPU预测，则需要在启动服务之前，请设置CUDA_VISIBLE_DEVICES环境变量，否则不用设置。
+
+## Step2: 发送预测请求
+
+配置好服务端，以下数行代码即可实现发送预测请求，获取预测结果
+
+```python
+import requests
+import json
+import cv2
+import base64
+
+import numpy as np
+
+
+def cv2_to_base64(image):
+    data = cv2.imencode('.jpg', image)[1]
+    return base64.b64encode(data.tostring()).decode('utf8')
+
+def base64_to_cv2(b64str):
+    data = base64.b64decode(b64str.encode('utf8'))
+    data = np.fromstring(data, np.uint8)
+    data = cv2.imdecode(data, cv2.IMREAD_COLOR)
+    return data
+
+# 发送HTTP请求
+org_im = cv2.imread('/PATH/TO/IMAGE')
+
+data = {'images':[cv2_to_base64(org_im)], 'top_k':2}
+headers = {"Content-type": "application/json"}
+url = "http://127.0.0.1:8866/predict/ghostnet_x1_0_imagenet"
+r = requests.post(url=url, headers=headers, data=json.dumps(data))
+data =r.json()["results"]['data']
+```
+
+### 查看代码
+
+[PaddleClas](https://github.com/PaddlePaddle/PaddleClas)
+
+### 依赖
+
+paddlepaddle >= 2.0.0
+
+paddlehub >= 2.0.0

modules/image/classification/ghostnet_x1_0_imagenet/label_list.txt

+tench
+goldfish
+great white shark
+tiger shark
+hammerhead
+electric ray
+stingray
+cock
+hen
+ostrich
+brambling
+goldfinch
+house finch
+junco
+indigo bunting
+robin
+bulbul
+jay
+magpie
+chickadee
+water ouzel
+kite
+bald eagle
+vulture
+great grey owl
+European fire salamander
+common newt
+eft
+spotted salamander
+axolotl
+bullfrog
+tree frog
+tailed frog
+loggerhead
+leatherback turtle
+mud turtle
+terrapin
+box turtle
+banded gecko
+common iguana
+American chameleon
+whiptail
+agama
+frilled lizard
+alligator lizard
+Gila monster
+green lizard
+African chameleon
+Komodo dragon
+African crocodile
+American alligator
+triceratops
+thunder snake
+ringneck snake
+hognose snake
+green snake
+king snake
+garter snake
+water snake
+vine snake
+night snake
+boa constrictor
+rock python
+Indian cobra
+green mamba
+sea snake
+horned viper
+diamondback
+sidewinder
+trilobite
+harvestman
+scorpion
+black and gold garden spider
+barn spider
+garden spider
+black widow
+tarantula
+wolf spider
+tick
+centipede
+black grouse
+ptarmigan
+ruffed grouse
+prairie chicken
+peacock
+quail
+partridge
+African grey
+macaw
+sulphur-crested cockatoo
+lorikeet
+coucal
+bee eater
+hornbill
+hummingbird
+jacamar
+toucan
+drake
+red-breasted merganser
+goose
+black swan
+tusker
+echidna
+platypus
+wallaby
+koala
+wombat
+jellyfish
+sea anemone
+brain coral
+flatworm
+nematode
+conch
+snail
+slug
+sea slug
+chiton
+chambered nautilus
+Dungeness crab
+rock crab
+fiddler crab
+king crab
+American lobster
+spiny lobster
+crayfish
+hermit crab
+isopod
+white stork
+black stork
+spoonbill
+flamingo
+little blue heron
+American egret
+bittern
+crane
+limpkin
+European gallinule
+American coot
+bustard
+ruddy turnstone
+red-backed sandpiper
+redshank
+dowitcher
+oystercatcher
+pelican
+king penguin
+albatross
+grey whale
+killer whale
+dugong
+sea lion
+Chihuahua
+Japanese spaniel
+Maltese dog
+Pekinese
+Shih-Tzu
+Blenheim spaniel
+papillon
+toy terrier
+Rhodesian ridgeback
+Afghan hound
+basset
+beagle
+bloodhound
+bluetick
+black-and-tan coonhound
+Walker hound
+English foxhound
+redbone
+borzoi
+Irish wolfhound
+Italian greyhound
+whippet
+Ibizan hound
+Norwegian elkhound
+otterhound
+Saluki
+Scottish deerhound
+Weimaraner
+Staffordshire bullterrier
+American Staffordshire terrier
+Bedlington terrier
+Border terrier
+Kerry blue terrier
+Irish terrier
+Norfolk terrier
+Norwich terrier
+Yorkshire terrier
+wire-haired fox terrier
+Lakeland terrier
+Sealyham terrier
+Airedale
+cairn
+Australian terrier
+Dandie Dinmont
+Boston bull
+miniature schnauzer
+giant schnauzer
+standard schnauzer
+Scotch terrier
+Tibetan terrier
+silky terrier
+soft-coated wheaten terrier
+West Highland white terrier
+Lhasa
+flat-coated retriever
+curly-coated retriever
+golden retriever
+Labrador retriever
+Chesapeake Bay retriever
+German short-haired pointer
+vizsla
+English setter
+Irish setter
+Gordon setter
+Brittany spaniel
+clumber
+English springer
+Welsh springer spaniel
+cocker spaniel
+Sussex spaniel
+Irish water spaniel
+kuvasz
+schipperke
+groenendael
+malinois
+briard
+kelpie
+komondor
+Old English sheepdog
+Shetland sheepdog
+collie
+Border collie
+Bouvier des Flandres
+Rottweiler
+German shepherd
+Doberman
+miniature pinscher
+Greater Swiss Mountain dog
+Bernese mountain dog
+Appenzeller
+EntleBucher
+boxer
+bull mastiff
+Tibetan mastiff
+French bulldog
+Great Dane
+Saint Bernard
+Eskimo dog
+malamute
+Siberian husky
+dalmatian
+affenpinscher
+basenji
+pug
+Leonberg
+Newfoundland
+Great Pyrenees
+Samoyed
+Pomeranian
+chow
+keeshond
+Brabancon griffon
+Pembroke
+Cardigan
+toy poodle
+miniature poodle
+standard poodle
+Mexican hairless
+timber wolf
+white wolf
+red wolf
+coyote
+dingo
+dhole
+African hunting dog
+hyena
+red fox
+kit fox
+Arctic fox
+grey fox
+tabby
+tiger cat
+Persian cat
+Siamese cat
+Egyptian cat
+cougar
+lynx
+leopard
+snow leopard
+jaguar
+lion
+tiger
+cheetah
+brown bear
+American black bear
+ice bear
+sloth bear
+mongoose
+meerkat
+tiger beetle
+ladybug
+ground beetle
+long-horned beetle
+leaf beetle
+dung beetle
+rhinoceros beetle
+weevil
+fly
+bee
+ant
+grasshopper
+cricket
+walking stick
+cockroach
+mantis
+cicada
+leafhopper
+lacewing
+dragonfly
+damselfly
+admiral
+ringlet
+monarch
+cabbage butterfly
+sulphur butterfly
+lycaenid
+starfish
+sea urchin
+sea cucumber
+wood rabbit
+hare
+Angora
+hamster
+porcupine
+fox squirrel
+marmot
+beaver
+guinea pig
+sorrel
+zebra
+hog
+wild boar
+warthog
+hippopotamus
+ox
+water buffalo
+bison
+ram
+bighorn
+ibex
+hartebeest
+impala
+gazelle
+Arabian camel
+llama
+weasel
+mink
+polecat
+black-footed ferret
+otter
+skunk
+badger
+armadillo
+three-toed sloth
+orangutan
+gorilla
+chimpanzee
+gibbon
+siamang
+guenon
+patas
+baboon
+macaque
+langur
+colobus
+proboscis monkey
+marmoset
+capuchin
+howler monkey
+titi
+spider monkey
+squirrel monkey
+Madagascar cat
+indri
+Indian elephant
+African elephant
+lesser panda
+giant panda
+barracouta
+eel
+coho
+rock beauty
+anemone fish
+sturgeon
+gar
+lionfish
+puffer
+abacus
+abaya
+academic gown
+accordion
+acoustic guitar
+aircraft carrier
+airliner
+airship
+altar
+ambulance
+amphibian
+analog clock
+apiary
+apron
+ashcan
+assault rifle
+backpack
+bakery
+balance beam
+balloon
+ballpoint
+Band Aid
+banjo
+bannister
+barbell
+barber chair
+barbershop
+barn
+barometer
+barrel
+barrow
+baseball
+basketball
+bassinet
+bassoon
+bathing cap
+bath towel
+bathtub
+beach wagon
+beacon
+beaker
+bearskin
+beer bottle
+beer glass
+bell cote
+bib
+bicycle-built-for-two
+bikini
+binder
+binoculars
+birdhouse
+boathouse
+bobsled
+bolo tie
+bonnet
+bookcase
+bookshop
+bottlecap
+bow
+bow tie
+brass
+brassiere
+breakwater
+breastplate
+broom
+bucket
+buckle
+bulletproof vest
+bullet train
+butcher shop
+cab
+caldron
+candle
+cannon
+canoe
+can opener
+cardigan
+car mirror
+carousel
+carpenters kit
+carton
+car wheel
+cash machine
+cassette
+cassette player
+castle
+catamaran
+CD player
+cello
+cellular telephone
+chain
+chainlink fence
+chain mail
+chain saw
+chest
+chiffonier
+chime
+china cabinet
+Christmas stocking
+church
+cinema
+cleaver
+cliff dwelling
+cloak
+clog
+cocktail shaker
+coffee mug
+coffeepot
+coil
+combination lock
+computer keyboard
+confectionery
+container ship
+convertible
+corkscrew
+cornet
+cowboy boot
+cowboy hat
+cradle
+crane
+crash helmet
+crate
+crib
+Crock Pot
+croquet ball
+crutch
+cuirass
+dam
+desk
+desktop computer
+dial telephone
+diaper
+digital clock
+digital watch
+dining table
+dishrag
+dishwasher
+disk brake
+dock
+dogsled
+dome
+doormat
+drilling platform
+drum
+drumstick
+dumbbell
+Dutch oven
+electric fan
+electric guitar
+electric locomotive
+entertainment center
+envelope
+espresso maker
+face powder
+feather boa
+file
+fireboat
+fire engine
+fire screen
+flagpole
+flute
+folding chair
+football helmet
+forklift
+fountain
+fountain pen
+four-poster
+freight car
+French horn
+frying pan
+fur coat
+garbage truck
+gasmask
+gas pump
+goblet
+go-kart
+golf ball
+golfcart
+gondola
+gong
+gown
+grand piano
+greenhouse
+grille
+grocery store
+guillotine
+hair slide
+hair spray
+half track
+hammer
+hamper
+hand blower
+hand-held computer
+handkerchief
+hard disc
+harmonica
+harp
+harvester
+hatchet
+holster
+home theater
+honeycomb
+hook
+hoopskirt
+horizontal bar
+horse cart
+hourglass
+iPod
+iron
+jack-o-lantern
+jean
+jeep
+jersey
+jigsaw puzzle
+jinrikisha
+joystick
+kimono
+knee pad
+knot
+lab coat
+ladle
+lampshade
+laptop
+lawn mower
+lens cap
+letter opener
+library
+lifeboat
+lighter
+limousine
+liner
+lipstick
+Loafer
+lotion
+loudspeaker
+loupe
+lumbermill
+magnetic compass
+mailbag
+mailbox
+maillot
+maillot
+manhole cover
+maraca
+marimba
+mask
+matchstick
+maypole
+maze
+measuring cup
+medicine chest
+megalith
+microphone
+microwave
+military uniform
+milk can
+minibus
+miniskirt
+minivan
+missile
+mitten
+mixing bowl
+mobile home
+Model T
+modem
+monastery
+monitor
+moped
+mortar
+mortarboard
+mosque
+mosquito net
+motor scooter
+mountain bike
+mountain tent
+mouse
+mousetrap
+moving van
+muzzle
+nail
+neck brace
+necklace
+nipple
+notebook
+obelisk
+oboe
+ocarina
+odometer
+oil filter
+organ
+oscilloscope
+overskirt
+oxcart
+oxygen mask
+packet
+paddle
+paddlewheel
+padlock
+paintbrush
+pajama
+palace
+panpipe
+paper towel
+parachute
+parallel bars
+park bench
+parking meter
+passenger car
+patio
+pay-phone
+pedestal
+pencil box
+pencil sharpener
+perfume
+Petri dish
+photocopier
+pick
+pickelhaube
+picket fence
+pickup
+pier
+piggy bank
+pill bottle
+pillow
+ping-pong ball
+pinwheel
+pirate
+pitcher
+plane
+planetarium
+plastic bag
+plate rack
+plow
+plunger
+Polaroid camera
+pole
+police van
+poncho
+pool table
+pop bottle
+pot
+potters wheel
+power drill
+prayer rug
+printer
+prison
+projectile
+projector
+puck
+punching bag
+purse
+quill
+quilt
+racer
+racket
+radiator
+radio
+radio telescope
+rain barrel
+recreational vehicle
+reel
+reflex camera
+refrigerator
+remote control
+restaurant
+revolver
+rifle
+rocking chair
+rotisserie
+rubber eraser
+rugby ball
+rule
+running shoe
+safe
+safety pin
+saltshaker
+sandal
+sarong
+sax
+scabbard
+scale
+school bus
+schooner
+scoreboard
+screen
+screw
+screwdriver
+seat belt
+sewing machine
+shield
+shoe shop
+shoji
+shopping basket
+shopping cart
+shovel
+shower cap
+shower curtain
+ski
+ski mask
+sleeping bag
+slide rule
+sliding door
+slot
+snorkel
+snowmobile
+snowplow
+soap dispenser
+soccer ball
+sock
+solar dish
+sombrero
+soup bowl
+space bar
+space heater
+space shuttle
+spatula
+speedboat
+spider web
+spindle
+sports car
+spotlight
+stage
+steam locomotive
+steel arch bridge
+steel drum
+stethoscope
+stole
+stone wall
+stopwatch
+stove
+strainer
+streetcar
+stretcher
+studio couch
+stupa
+submarine
+suit
+sundial
+sunglass
+sunglasses
+sunscreen
+suspension bridge
+swab
+sweatshirt
+swimming trunks
+swing
+switch
+syringe
+table lamp
+tank
+tape player
+teapot
+teddy
+television
+tennis ball
+thatch
+theater curtain
+thimble
+thresher
+throne
+tile roof
+toaster
+tobacco shop
+toilet seat
+torch
+totem pole
+tow truck
+toyshop
+tractor
+trailer truck
+tray
+trench coat
+tricycle
+trimaran
+tripod
+triumphal arch
+trolleybus
+trombone
+tub
+turnstile
+typewriter keyboard
+umbrella
+unicycle
+upright
+vacuum
+vase
+vault
+velvet
+vending machine
+vestment
+viaduct
+violin
+volleyball
+waffle iron
+wall clock
+wallet
+wardrobe
+warplane
+washbasin
+washer
+water bottle
+water jug
+water tower
+whiskey jug
+whistle
+wig
+window screen
+window shade
+Windsor tie
+wine bottle
+wing
+wok
+wooden spoon
+wool
+worm fence
+wreck
+yawl
+yurt
+web site
+comic book
+crossword puzzle
+street sign
+traffic light
+book jacket
+menu
+plate
+guacamole
+consomme
+hot pot
+trifle
+ice cream
+ice lolly
+French loaf
+bagel
+pretzel
+cheeseburger
+hotdog
+mashed potato
+head cabbage
+broccoli
+cauliflower
+zucchini
+spaghetti squash
+acorn squash
+butternut squash
+cucumber
+artichoke
+bell pepper
+cardoon
+mushroom
+Granny Smith
+strawberry
+orange
+lemon
+fig
+pineapple
+banana
+jackfruit
+custard apple
+pomegranate
+hay
+carbonara
+chocolate sauce
+dough
+meat loaf
+pizza
+potpie
+burrito
+red wine
+espresso
+cup
+eggnog
+alp
+bubble
+cliff
+coral reef
+geyser
+lakeside
+promontory
+sandbar
+seashore
+valley
+volcano
+ballplayer
+groom
+scuba diver
+rapeseed
+daisy
+yellow ladys slipper
+corn
+acorn
+hip
+buckeye
+coral fungus
+agaric
+gyromitra
+stinkhorn
+earthstar
+hen-of-the-woods
+bolete
+ear
+toilet tissue

modules/image/classification/ghostnet_x1_0_imagenet/module.py

+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import math
+from typing import Union
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+import paddlehub.vision.transforms as T
+import numpy as np
+from paddle import ParamAttr
+from paddle.nn.initializer import Uniform, KaimingNormal
+from paddlehub.module.module import moduleinfo
+from paddlehub.module.cv_module import ImageClassifierModule
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(self, in_channels, out_channels, kernel_size, stride=1, groups=1, act="relu", name=None):
+        super(ConvBNLayer, self).__init__()
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal(), name=name + "_weights"),
+            bias_attr=False)
+        bn_name = name + "_bn"
+
+        self._batch_norm = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale", regularizer=paddle.regularizer.L2Decay(0.0)),
+            bias_attr=ParamAttr(name=bn_name + "_offset", regularizer=paddle.regularizer.L2Decay(0.0)),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance")
+
+    def forward(self, inputs):
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class SEBlock(nn.Layer):
+    def __init__(self, num_channels, reduction_ratio=4, name=None):
+        super(SEBlock, self).__init__()
+        self.pool2d_gap = nn.AdaptiveAvgPool2D(1)
+        self._num_channels = num_channels
+        stdv = 1.0 / math.sqrt(num_channels * 1.0)
+        med_ch = num_channels // reduction_ratio
+        self.squeeze = nn.Linear(
+            num_channels,
+            med_ch,
+            weight_attr=ParamAttr(initializer=Uniform(-stdv, stdv), name=name + "_1_weights"),
+            bias_attr=ParamAttr(name=name + "_1_offset"))
+        stdv = 1.0 / math.sqrt(med_ch * 1.0)
+        self.excitation = nn.Linear(
+            med_ch,
+            num_channels,
+            weight_attr=ParamAttr(initializer=Uniform(-stdv, stdv), name=name + "_2_weights"),
+            bias_attr=ParamAttr(name=name + "_2_offset"))
+
+    def forward(self, inputs):
+        pool = self.pool2d_gap(inputs)
+        pool = paddle.squeeze(pool, axis=[2, 3])
+        squeeze = self.squeeze(pool)
+        squeeze = F.relu(squeeze)
+        excitation = self.excitation(squeeze)
+        excitation = paddle.clip(x=excitation, min=0, max=1)
+        excitation = paddle.unsqueeze(excitation, axis=[2, 3])
+        out = paddle.multiply(inputs, excitation)
+        return out
+
+
+class GhostModule(nn.Layer):
+    def __init__(self, in_channels, output_channels, kernel_size=1, ratio=2, dw_size=3, stride=1, relu=True, name=None):
+        super(GhostModule, self).__init__()
+        init_channels = int(math.ceil(output_channels / ratio))
+        new_channels = int(init_channels * (ratio - 1))
+        self.primary_conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=init_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            groups=1,
+            act="relu" if relu else None,
+            name=name + "_primary_conv")
+        self.cheap_operation = ConvBNLayer(
+            in_channels=init_channels,
+            out_channels=new_channels,
+            kernel_size=dw_size,
+            stride=1,
+            groups=init_channels,
+            act="relu" if relu else None,
+            name=name + "_cheap_operation")
+
+    def forward(self, inputs):
+        x = self.primary_conv(inputs)
+        y = self.cheap_operation(x)
+        out = paddle.concat([x, y], axis=1)
+        return out
+
+
+class GhostBottleneck(nn.Layer):
+    def __init__(self, in_channels, hidden_dim, output_channels, kernel_size, stride, use_se, name=None):
+        super(GhostBottleneck, self).__init__()
+        self._stride = stride
+        self._use_se = use_se
+        self._num_channels = in_channels
+        self._output_channels = output_channels
+        self.ghost_module_1 = GhostModule(
+            in_channels=in_channels,
+            output_channels=hidden_dim,
+            kernel_size=1,
+            stride=1,
+            relu=True,
+            name=name + "_ghost_module_1")
+        if stride == 2:
+            self.depthwise_conv = ConvBNLayer(
+                in_channels=hidden_dim,
+                out_channels=hidden_dim,
+                kernel_size=kernel_size,
+                stride=stride,
+                groups=hidden_dim,
+                act=None,
+                name=name + "_depthwise_depthwise"  # looks strange due to an old typo, will be fixed later.
+            )
+        if use_se:
+            self.se_block = SEBlock(num_channels=hidden_dim, name=name + "_se")
+        self.ghost_module_2 = GhostModule(
+            in_channels=hidden_dim,
+            output_channels=output_channels,
+            kernel_size=1,
+            relu=False,
+            name=name + "_ghost_module_2")
+        if stride != 1 or in_channels != output_channels:
+            self.shortcut_depthwise = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=in_channels,
+                kernel_size=kernel_size,
+                stride=stride,
+                groups=in_channels,
+                act=None,
+                name=name + "_shortcut_depthwise_depthwise"  # looks strange due to an old typo, will be fixed later.
+            )
+            self.shortcut_conv = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=output_channels,
+                kernel_size=1,
+                stride=1,
+                groups=1,
+                act=None,
+                name=name + "_shortcut_conv")
+
+    def forward(self, inputs):
+        x = self.ghost_module_1(inputs)
+        if self._stride == 2:
+            x = self.depthwise_conv(x)
+        if self._use_se:
+            x = self.se_block(x)
+        x = self.ghost_module_2(x)
+        if self._stride == 1 and self._num_channels == self._output_channels:
+            shortcut = inputs
+        else:
+            shortcut = self.shortcut_depthwise(inputs)
+            shortcut = self.shortcut_conv(shortcut)
+        return paddle.add(x=x, y=shortcut)
+
+
+@moduleinfo(
+    name="ghostnet_x1_0_imagenet",
+    type="CV/classification",
+    author="paddlepaddle",
+    author_email="",
+    summary="ghostnet_x1_0_imagenet is a classification model, "
+    "this module is trained with Imagenet dataset.",
+    version="1.0.0",
+    meta=ImageClassifierModule)
+class GhostNet(nn.Layer):
+    def __init__(self, label_list: list = None, load_checkpoint: str = None):
+        super(GhostNet, self).__init__()
+
+        if label_list is not None:
+            self.labels = label_list
+            class_dim = len(self.labels)
+        else:
+            label_list = []
+            label_file = os.path.join(self.directory, 'label_list.txt')
+            files = open(label_file)
+            for line in files.readlines():
+                line = line.strip('\n')
+                label_list.append(line)
+            self.labels = label_list
+            class_dim = len(self.labels)
+
+        self.cfgs = [
+            # k, t, c, SE, s
+            [3, 16, 16, 0, 1],
+            [3, 48, 24, 0, 2],
+            [3, 72, 24, 0, 1],
+            [5, 72, 40, 1, 2],
+            [5, 120, 40, 1, 1],
+            [3, 240, 80, 0, 2],
+            [3, 200, 80, 0, 1],
+            [3, 184, 80, 0, 1],
+            [3, 184, 80, 0, 1],
+            [3, 480, 112, 1, 1],
+            [3, 672, 112, 1, 1],
+            [5, 672, 160, 1, 2],
+            [5, 960, 160, 0, 1],
+            [5, 960, 160, 1, 1],
+            [5, 960, 160, 0, 1],
+            [5, 960, 160, 1, 1]
+        ]
+        self.scale = 1.0
+        output_channels = int(self._make_divisible(16 * self.scale, 4))
+        self.conv1 = ConvBNLayer(
+            in_channels=3, out_channels=output_channels, kernel_size=3, stride=2, groups=1, act="relu", name="conv1")
+        # build inverted residual blocks
+        idx = 0
+        self.ghost_bottleneck_list = []
+        for k, exp_size, c, use_se, s in self.cfgs:
+            in_channels = output_channels
+            output_channels = int(self._make_divisible(c * self.scale, 4))
+            hidden_dim = int(self._make_divisible(exp_size * self.scale, 4))
+            ghost_bottleneck = self.add_sublayer(
+                name="_ghostbottleneck_" + str(idx),
+                sublayer=GhostBottleneck(
+                    in_channels=in_channels,
+                    hidden_dim=hidden_dim,
+                    output_channels=output_channels,
+                    kernel_size=k,
+                    stride=s,
+                    use_se=use_se,
+                    name="_ghostbottleneck_" + str(idx)))
+            self.ghost_bottleneck_list.append(ghost_bottleneck)
+            idx += 1
+        # build last several layers
+        in_channels = output_channels
+        output_channels = int(self._make_divisible(exp_size * self.scale, 4))
+        self.conv_last = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=output_channels,
+            kernel_size=1,
+            stride=1,
+            groups=1,
+            act="relu",
+            name="conv_last")
+        self.pool2d_gap = nn.AdaptiveAvgPool2D(1)
+        in_channels = output_channels
+        self._fc0_output_channels = 1280
+        self.fc_0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=self._fc0_output_channels,
+            kernel_size=1,
+            stride=1,
+            act="relu",
+            name="fc_0")
+        self.dropout = nn.Dropout(p=0.2)
+        stdv = 1.0 / math.sqrt(self._fc0_output_channels * 1.0)
+        self.fc_1 = nn.Linear(
+            self._fc0_output_channels,
+            class_dim,
+            weight_attr=ParamAttr(name="fc_1_weights", initializer=Uniform(-stdv, stdv)),
+            bias_attr=ParamAttr(name="fc_1_offset"))
+
+        if load_checkpoint is not None:
+            self.model_dict = paddle.load(load_checkpoint)
+            self.set_dict(self.model_dict)
+            print("load custom checkpoint success")
+        else:
+            checkpoint = os.path.join(self.directory, 'model.pdparams')
+            self.model_dict = paddle.load(checkpoint)
+            self.set_dict(self.model_dict)
+            print("load pretrained checkpoint success")
+
+    def transforms(self, images: Union[str, np.ndarray]):
+        transforms = T.Compose([
+            T.Resize((256, 256)),
+            T.CenterCrop(224),
+            T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+        ],
+                               to_rgb=True)
+        return transforms(images).astype('float32')
+
+    def forward(self, inputs):
+        x = self.conv1(inputs)
+        for ghost_bottleneck in self.ghost_bottleneck_list:
+            x = ghost_bottleneck(x)
+        x = self.conv_last(x)
+        feature = self.pool2d_gap(x)
+        x = self.fc_0(feature)
+        x = self.dropout(x)
+        x = paddle.reshape(x, shape=[-1, self._fc0_output_channels])
+        x = self.fc_1(x)
+        return x, feature
+
+    def _make_divisible(self, v, divisor, min_value=None):
+        """
+        This function is taken from the original tf repo.
+        It ensures that all layers have a channel number that is divisible by 8
+        It can be seen here:
+        https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
+        """
+        if min_value is None:
+            min_value = divisor
+        new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+        # Make sure that round down does not go down by more than 10%.
+        if new_v < 0.9 * v:
+            new_v += divisor
+        return new_v

modules/image/classification/ghostnet_x1_3_imagenet/README.md

+```shell
+$ hub install ghostnet_x1_3_imagenet==1.0.0
+```
+
+## 命令行预测
+
+```shell
+$ hub run ghostnet_x1_3_imagenet --input_path "/PATH/TO/IMAGE" --top_k 5
+```
+
+## 脚本预测
+
+```python
+import paddle
+import paddlehub as hub
+
+if __name__ == '__main__':
+
+    model = hub.Module(name='ghostnet_x1_3_imagenet',)
+    result = model.predict([PATH/TO/IMAGE])
+```
+
+## Fine-tune代码步骤
+
+使用PaddleHub Fine-tune API进行Fine-tune可以分为4个步骤。
+
+### Step1: 定义数据预处理方式
+```python
+import paddlehub.vision.transforms as T
+
+transforms = T.Compose([T.Resize((256, 256)),
+                        T.CenterCrop(224),
+                        T.Normalize(mean=[0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225])],
+                        to_rgb=True)
+```
+
+'transforms' 数据增强模块定义了丰富的数据预处理方式，用户可按照需求替换自己需要的数据预处理方式。
+
+### Step2: 下载数据集并使用
+```python
+from paddlehub.datasets import Flowers
+
+flowers = Flowers(transforms)
+flowers_validate = Flowers(transforms, mode='val')
+```
+* transforms(Callable): 数据预处理方式。
+* mode(str): 选择数据模式，可选项有 'train', 'test', 'val'， 默认为'train'。
+
+'hub.datasets.Flowers()' 会自动从网络下载数据集并解压到用户目录下'$HOME/.paddlehub/dataset'目录。
+
+
+### Step3: 加载预训练模型
+
+```python
+import paddlehub as hub
+
+model = hub.Module(name='ghostnet_x1_3_imagenet',
+                   label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"],
+                   load_checkpoint=None)
+```
+* name(str): 选择预训练模型的名字。
+* label_list(list): 设置标签对应分类类别, 默认为Imagenet2012类别。
+* load _checkpoint(str): 模型参数地址。
+
+PaddleHub提供许多图像分类预训练模型，如xception、mobilenet、efficientnet等，详细信息参见[图像分类模型](https://www.paddlepaddle.org.cn/hub?filter=en_category&value=ImageClassification)。
+
+如果想尝试efficientnet模型，只需要更换Module中的'name'参数即可.
+```python
+import paddlehub as hub
+
+# 更换name参数即可无缝切换efficientnet模型, 代码示例如下
+module = hub.Module(name="efficientnetb7_imagenet")
+```
+**NOTE:**目前部分模型还没有完全升级到2.0版本，敬请期待。
+
+### Step4: 选择优化策略和运行配置
+
+```python
+import paddle
+from paddlehub.finetune.trainer import Trainer
+
+optimizer = paddle.optimizer.Adam(learning_rate=0.001, parameters=model.parameters())
+trainer = Trainer(model, optimizer, checkpoint_dir='img_classification_ckpt')
+
+trainer.train(flowers, epochs=100, batch_size=32, eval_dataset=flowers_validate, save_interval=1)
+```
+
+#### 优化策略
+
+Paddle2.0rc提供了多种优化器选择，如'SGD', 'Adam', 'Adamax'等，详细参见[策略](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/optimizer/optimizer/Optimizer_cn.html)。
+
+其中'Adam':
+
+* learning_rate: 全局学习率。默认为1e-3；
+* parameters: 待优化模型参数。
+
+#### 运行配置
+'Trainer' 主要控制Fine-tune的训练，包含以下可控制的参数:
+
+* model: 被优化模型；
+* optimizer: 优化器选择；
+* use_vdl: 是否使用vdl可视化训练过程；
+* checkpoint_dir: 保存模型参数的地址；
+* compare_metrics: 保存最优模型的衡量指标；
+
+'trainer.train' 主要控制具体的训练过程，包含以下可控制的参数：
+
+* train_dataset: 训练时所用的数据集；
+* epochs: 训练轮数；
+* batch_size: 训练的批大小，如果使用GPU，请根据实际情况调整batch_size；
+* num_workers: works的数量，默认为0；
+* eval_dataset: 验证集；
+* log_interval: 打印日志的间隔， 单位为执行批训练的次数。
+* save_interval: 保存模型的间隔频次，单位为执行训练的轮数。
+
+## 模型预测
+
+当完成Fine-tune后，Fine-tune过程在验证集上表现最优的模型会被保存在'${CHECKPOINT_DIR}/best_model'目录下，其中'${CHECKPOINT_DIR}'目录为Fine-tune时所选择的保存checkpoint的目录。
+我们使用该模型来进行预测。predict.py脚本如下：
+
+```python
+import paddle
+import paddlehub as hub
+
+if __name__ == '__main__':
+
+    model = hub.Module(name='ghostnet_x1_3_imagenet', label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"], load_checkpoint='/PATH/TO/CHECKPOINT')
+    result = model.predict(['flower.jpg'])
+```
+
+参数配置正确后，请执行脚本'python predict.py'， 加载模型具体可参见[加载](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/framework/io/load_cn.html#load)。
+
+**NOTE:** 进行预测时，所选择的module，checkpoint_dir，dataset必须和Fine-tune所用的一样。
+
+## 服务部署
+
+PaddleHub Serving可以部署一个在线分类任务服务
+
+## Step1: 启动PaddleHub Serving
+
+运行启动命令：
+
+```shell
+$ hub serving start -m ghostnet_x1_3_imagenet
+```
+
+这样就完成了一个分类任务服务化API的部署，默认端口号为8866。
+
+**NOTE:** 如使用GPU预测，则需要在启动服务之前，请设置CUDA_VISIBLE_DEVICES环境变量，否则不用设置。
+
+## Step2: 发送预测请求
+
+配置好服务端，以下数行代码即可实现发送预测请求，获取预测结果
+
+```python
+import requests
+import json
+import cv2
+import base64
+
+import numpy as np
+
+
+def cv2_to_base64(image):
+    data = cv2.imencode('.jpg', image)[1]
+    return base64.b64encode(data.tostring()).decode('utf8')
+
+def base64_to_cv2(b64str):
+    data = base64.b64decode(b64str.encode('utf8'))
+    data = np.fromstring(data, np.uint8)
+    data = cv2.imdecode(data, cv2.IMREAD_COLOR)
+    return data
+
+# 发送HTTP请求
+org_im = cv2.imread('/PATH/TO/IMAGE')
+
+data = {'images':[cv2_to_base64(org_im)], 'top_k':2}
+headers = {"Content-type": "application/json"}
+url = "http://127.0.0.1:8866/predict/ghostnet_x1_3_imagenet"
+r = requests.post(url=url, headers=headers, data=json.dumps(data))
+data =r.json()["results"]['data']
+```
+
+### 查看代码
+
+[PaddleClas](https://github.com/PaddlePaddle/PaddleClas)
+
+### 依赖
+
+paddlepaddle >= 2.0.0
+
+paddlehub >= 2.0.0

modules/image/classification/ghostnet_x1_3_imagenet/module.py

+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import math
+from typing import Union
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+import paddlehub.vision.transforms as T
+import numpy as np
+from paddle import ParamAttr
+from paddle.nn.initializer import Uniform, KaimingNormal
+from paddlehub.module.module import moduleinfo
+from paddlehub.module.cv_module import ImageClassifierModule
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(self, in_channels, out_channels, kernel_size, stride=1, groups=1, act="relu", name=None):
+        super(ConvBNLayer, self).__init__()
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal(), name=name + "_weights"),
+            bias_attr=False)
+        bn_name = name + "_bn"
+
+        self._batch_norm = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale", regularizer=paddle.regularizer.L2Decay(0.0)),
+            bias_attr=ParamAttr(name=bn_name + "_offset", regularizer=paddle.regularizer.L2Decay(0.0)),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance")
+
+    def forward(self, inputs):
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class SEBlock(nn.Layer):
+    def __init__(self, num_channels, reduction_ratio=4, name=None):
+        super(SEBlock, self).__init__()
+        self.pool2d_gap = nn.AdaptiveAvgPool2D(1)
+        self._num_channels = num_channels
+        stdv = 1.0 / math.sqrt(num_channels * 1.0)
+        med_ch = num_channels // reduction_ratio
+        self.squeeze = nn.Linear(
+            num_channels,
+            med_ch,
+            weight_attr=ParamAttr(initializer=Uniform(-stdv, stdv), name=name + "_1_weights"),
+            bias_attr=ParamAttr(name=name + "_1_offset"))
+        stdv = 1.0 / math.sqrt(med_ch * 1.0)
+        self.excitation = nn.Linear(
+            med_ch,
+            num_channels,
+            weight_attr=ParamAttr(initializer=Uniform(-stdv, stdv), name=name + "_2_weights"),
+            bias_attr=ParamAttr(name=name + "_2_offset"))
+
+    def forward(self, inputs):
+        pool = self.pool2d_gap(inputs)
+        pool = paddle.squeeze(pool, axis=[2, 3])
+        squeeze = self.squeeze(pool)
+        squeeze = F.relu(squeeze)
+        excitation = self.excitation(squeeze)
+        excitation = paddle.clip(x=excitation, min=0, max=1)
+        excitation = paddle.unsqueeze(excitation, axis=[2, 3])
+        out = paddle.multiply(inputs, excitation)
+        return out
+
+
+class GhostModule(nn.Layer):
+    def __init__(self, in_channels, output_channels, kernel_size=1, ratio=2, dw_size=3, stride=1, relu=True, name=None):
+        super(GhostModule, self).__init__()
+        init_channels = int(math.ceil(output_channels / ratio))
+        new_channels = int(init_channels * (ratio - 1))
+        self.primary_conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=init_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            groups=1,
+            act="relu" if relu else None,
+            name=name + "_primary_conv")
+        self.cheap_operation = ConvBNLayer(
+            in_channels=init_channels,
+            out_channels=new_channels,
+            kernel_size=dw_size,
+            stride=1,
+            groups=init_channels,
+            act="relu" if relu else None,
+            name=name + "_cheap_operation")
+
+    def forward(self, inputs):
+        x = self.primary_conv(inputs)
+        y = self.cheap_operation(x)
+        out = paddle.concat([x, y], axis=1)
+        return out
+
+
+class GhostBottleneck(nn.Layer):
+    def __init__(self, in_channels, hidden_dim, output_channels, kernel_size, stride, use_se, name=None):
+        super(GhostBottleneck, self).__init__()
+        self._stride = stride
+        self._use_se = use_se
+        self._num_channels = in_channels
+        self._output_channels = output_channels
+        self.ghost_module_1 = GhostModule(
+            in_channels=in_channels,
+            output_channels=hidden_dim,
+            kernel_size=1,
+            stride=1,
+            relu=True,
+            name=name + "_ghost_module_1")
+        if stride == 2:
+            self.depthwise_conv = ConvBNLayer(
+                in_channels=hidden_dim,
+                out_channels=hidden_dim,
+                kernel_size=kernel_size,
+                stride=stride,
+                groups=hidden_dim,
+                act=None,
+                name=name + "_depthwise_depthwise"  # looks strange due to an old typo, will be fixed later.
+            )
+        if use_se:
+            self.se_block = SEBlock(num_channels=hidden_dim, name=name + "_se")
+        self.ghost_module_2 = GhostModule(
+            in_channels=hidden_dim,
+            output_channels=output_channels,
+            kernel_size=1,
+            relu=False,
+            name=name + "_ghost_module_2")
+        if stride != 1 or in_channels != output_channels:
+            self.shortcut_depthwise = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=in_channels,
+                kernel_size=kernel_size,
+                stride=stride,
+                groups=in_channels,
+                act=None,
+                name=name + "_shortcut_depthwise_depthwise"  # looks strange due to an old typo, will be fixed later.
+            )
+            self.shortcut_conv = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=output_channels,
+                kernel_size=1,
+                stride=1,
+                groups=1,
+                act=None,
+                name=name + "_shortcut_conv")
+
+    def forward(self, inputs):
+        x = self.ghost_module_1(inputs)
+        if self._stride == 2:
+            x = self.depthwise_conv(x)
+        if self._use_se:
+            x = self.se_block(x)
+        x = self.ghost_module_2(x)
+        if self._stride == 1 and self._num_channels == self._output_channels:
+            shortcut = inputs
+        else:
+            shortcut = self.shortcut_depthwise(inputs)
+            shortcut = self.shortcut_conv(shortcut)
+        return paddle.add(x=x, y=shortcut)
+
+
+@moduleinfo(
+    name="ghostnet_x1_3_imagenet",
+    type="CV/classification",
+    author="paddlepaddle",
+    author_email="",
+    summary="ghostnet_x1_3_imagenet is a classification model, "
+    "this module is trained with Imagenet dataset.",
+    version="1.0.0",
+    meta=ImageClassifierModule)
+class GhostNet(nn.Layer):
+    def __init__(self, label_list: list = None, load_checkpoint: str = None):
+        super(GhostNet, self).__init__()
+
+        if label_list is not None:
+            self.labels = label_list
+            class_dim = len(self.labels)
+        else:
+            label_list = []
+            label_file = os.path.join(self.directory, 'label_list.txt')
+            files = open(label_file)
+            for line in files.readlines():
+                line = line.strip('\n')
+                label_list.append(line)
+            self.labels = label_list
+            class_dim = len(self.labels)
+
+        self.cfgs = [
+            # k, t, c, SE, s
+            [3, 16, 16, 0, 1],
+            [3, 48, 24, 0, 2],
+            [3, 72, 24, 0, 1],
+            [5, 72, 40, 1, 2],
+            [5, 120, 40, 1, 1],
+            [3, 240, 80, 0, 2],
+            [3, 200, 80, 0, 1],
+            [3, 184, 80, 0, 1],
+            [3, 184, 80, 0, 1],
+            [3, 480, 112, 1, 1],
+            [3, 672, 112, 1, 1],
+            [5, 672, 160, 1, 2],
+            [5, 960, 160, 0, 1],
+            [5, 960, 160, 1, 1],
+            [5, 960, 160, 0, 1],
+            [5, 960, 160, 1, 1]
+        ]
+        self.scale = 1.3
+        output_channels = int(self._make_divisible(16 * self.scale, 4))
+        self.conv1 = ConvBNLayer(
+            in_channels=3, out_channels=output_channels, kernel_size=3, stride=2, groups=1, act="relu", name="conv1")
+        # build inverted residual blocks
+        idx = 0
+        self.ghost_bottleneck_list = []
+        for k, exp_size, c, use_se, s in self.cfgs:
+            in_channels = output_channels
+            output_channels = int(self._make_divisible(c * self.scale, 4))
+            hidden_dim = int(self._make_divisible(exp_size * self.scale, 4))
+            ghost_bottleneck = self.add_sublayer(
+                name="_ghostbottleneck_" + str(idx),
+                sublayer=GhostBottleneck(
+                    in_channels=in_channels,
+                    hidden_dim=hidden_dim,
+                    output_channels=output_channels,
+                    kernel_size=k,
+                    stride=s,
+                    use_se=use_se,
+                    name="_ghostbottleneck_" + str(idx)))
+            self.ghost_bottleneck_list.append(ghost_bottleneck)
+            idx += 1
+        # build last several layers
+        in_channels = output_channels
+        output_channels = int(self._make_divisible(exp_size * self.scale, 4))
+        self.conv_last = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=output_channels,
+            kernel_size=1,
+            stride=1,
+            groups=1,
+            act="relu",
+            name="conv_last")
+        self.pool2d_gap = nn.AdaptiveAvgPool2D(1)
+        in_channels = output_channels
+        self._fc0_output_channels = 1280
+        self.fc_0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=self._fc0_output_channels,
+            kernel_size=1,
+            stride=1,
+            act="relu",
+            name="fc_0")
+        self.dropout = nn.Dropout(p=0.2)
+        stdv = 1.0 / math.sqrt(self._fc0_output_channels * 1.0)
+        self.fc_1 = nn.Linear(
+            self._fc0_output_channels,
+            class_dim,
+            weight_attr=ParamAttr(name="fc_1_weights", initializer=Uniform(-stdv, stdv)),
+            bias_attr=ParamAttr(name="fc_1_offset"))
+
+        if load_checkpoint is not None:
+            self.model_dict = paddle.load(load_checkpoint)
+            self.set_dict(self.model_dict)
+            print("load custom checkpoint success")
+        else:
+            checkpoint = os.path.join(self.directory, 'model.pdparams')
+            self.model_dict = paddle.load(checkpoint)
+            self.set_dict(self.model_dict)
+            print("load pretrained checkpoint success")
+
+    def transforms(self, images: Union[str, np.ndarray]):
+        transforms = T.Compose([
+            T.Resize((256, 256)),
+            T.CenterCrop(224),
+            T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+        ],
+                               to_rgb=True)
+        return transforms(images).astype('float32')
+
+    def forward(self, inputs):
+        x = self.conv1(inputs)
+        for ghost_bottleneck in self.ghost_bottleneck_list:
+            x = ghost_bottleneck(x)
+        x = self.conv_last(x)
+        feature = self.pool2d_gap(x)
+        x = self.fc_0(feature)
+        x = self.dropout(x)
+        x = paddle.reshape(x, shape=[-1, self._fc0_output_channels])
+        x = self.fc_1(x)
+        return x, feature
+
+    def _make_divisible(self, v, divisor, min_value=None):
+        """
+        This function is taken from the original tf repo.
+        It ensures that all layers have a channel number that is divisible by 8
+        It can be seen here:
+        https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
+        """
+        if min_value is None:
+            min_value = divisor
+        new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+        # Make sure that round down does not go down by more than 10%.
+        if new_v < 0.9 * v:
+            new_v += divisor
+        return new_v

modules/image/classification/ghostnet_x1_3_imagenet_ssld/README.md

+```shell
+$ hub install ghostnet_x1_3_imagenet_ssld==1.0.0
+```
+
+## 命令行预测
+
+```shell
+$ hub run ghostnet_x1_3_imagenet_ssld --input_path "/PATH/TO/IMAGE" --top_k 5
+```
+
+## 脚本预测
+
+```python
+import paddle
+import paddlehub as hub
+
+if __name__ == '__main__':
+
+    model = hub.Module(name='ghostnet_x1_3_imagenet_ssld',)
+    result = model.predict([PATH/TO/IMAGE])
+```
+
+## Fine-tune代码步骤
+
+使用PaddleHub Fine-tune API进行Fine-tune可以分为4个步骤。
+
+### Step1: 定义数据预处理方式
+```python
+import paddlehub.vision.transforms as T
+
+transforms = T.Compose([T.Resize((256, 256)),
+                        T.CenterCrop(224),
+                        T.Normalize(mean=[0.485, 0.456, 0.406], std = [0.229, 0.224, 0.225])],
+                        to_rgb=True)
+```
+
+'transforms' 数据增强模块定义了丰富的数据预处理方式，用户可按照需求替换自己需要的数据预处理方式。
+
+### Step2: 下载数据集并使用
+```python
+from paddlehub.datasets import Flowers
+
+flowers = Flowers(transforms)
+flowers_validate = Flowers(transforms, mode='val')
+```
+* transforms(Callable): 数据预处理方式。
+* mode(str): 选择数据模式，可选项有 'train', 'test', 'val'， 默认为'train'。
+
+'hub.datasets.Flowers()' 会自动从网络下载数据集并解压到用户目录下'$HOME/.paddlehub/dataset'目录。
+
+
+### Step3: 加载预训练模型
+
+```python
+import paddlehub as hub
+
+model = hub.Module(name='ghostnet_x1_3_imagenet_ssld',
+                   label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"],
+                   load_checkpoint=None)
+```
+* name(str): 选择预训练模型的名字。
+* label_list(list): 设置标签对应分类类别, 默认为Imagenet2012类别。
+* load _checkpoint(str): 模型参数地址。
+
+PaddleHub提供许多图像分类预训练模型，如xception、mobilenet、efficientnet等，详细信息参见[图像分类模型](https://www.paddlepaddle.org.cn/hub?filter=en_category&value=ImageClassification)。
+
+如果想尝试efficientnet模型，只需要更换Module中的'name'参数即可.
+```python
+import paddlehub as hub
+
+# 更换name参数即可无缝切换efficientnet模型, 代码示例如下
+module = hub.Module(name="efficientnetb7_imagenet")
+```
+**NOTE:**目前部分模型还没有完全升级到2.0版本，敬请期待。
+
+### Step4: 选择优化策略和运行配置
+
+```python
+import paddle
+from paddlehub.finetune.trainer import Trainer
+
+optimizer = paddle.optimizer.Adam(learning_rate=0.001, parameters=model.parameters())
+trainer = Trainer(model, optimizer, checkpoint_dir='img_classification_ckpt')
+
+trainer.train(flowers, epochs=100, batch_size=32, eval_dataset=flowers_validate, save_interval=1)
+```
+
+#### 优化策略
+
+Paddle2.0rc提供了多种优化器选择，如'SGD', 'Adam', 'Adamax'等，详细参见[策略](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/optimizer/optimizer/Optimizer_cn.html)。
+
+其中'Adam':
+
+* learning_rate: 全局学习率。默认为1e-3；
+* parameters: 待优化模型参数。
+
+#### 运行配置
+'Trainer' 主要控制Fine-tune的训练，包含以下可控制的参数:
+
+* model: 被优化模型；
+* optimizer: 优化器选择；
+* use_vdl: 是否使用vdl可视化训练过程；
+* checkpoint_dir: 保存模型参数的地址；
+* compare_metrics: 保存最优模型的衡量指标；
+
+'trainer.train' 主要控制具体的训练过程，包含以下可控制的参数：
+
+* train_dataset: 训练时所用的数据集；
+* epochs: 训练轮数；
+* batch_size: 训练的批大小，如果使用GPU，请根据实际情况调整batch_size；
+* num_workers: works的数量，默认为0；
+* eval_dataset: 验证集；
+* log_interval: 打印日志的间隔， 单位为执行批训练的次数。
+* save_interval: 保存模型的间隔频次，单位为执行训练的轮数。
+
+## 模型预测
+
+当完成Fine-tune后，Fine-tune过程在验证集上表现最优的模型会被保存在'${CHECKPOINT_DIR}/best_model'目录下，其中'${CHECKPOINT_DIR}'目录为Fine-tune时所选择的保存checkpoint的目录。
+我们使用该模型来进行预测。predict.py脚本如下：
+
+```python
+import paddle
+import paddlehub as hub
+
+if __name__ == '__main__':
+
+    model = hub.Module(name='ghostnet_x1_3_imagenet_ssld', label_list=["roses", "tulips", "daisy", "sunflowers", "dandelion"], load_checkpoint='/PATH/TO/CHECKPOINT')
+    result = model.predict(['flower.jpg'])
+```
+
+参数配置正确后，请执行脚本'python predict.py'， 加载模型具体可参见[加载](https://www.paddlepaddle.org.cn/documentation/docs/zh/2.0-rc/api/paddle/framework/io/load_cn.html#load)。
+
+**NOTE:** 进行预测时，所选择的module，checkpoint_dir，dataset必须和Fine-tune所用的一样。
+
+## 服务部署
+
+PaddleHub Serving可以部署一个在线分类任务服务
+
+## Step1: 启动PaddleHub Serving
+
+运行启动命令：
+
+```shell
+$ hub serving start -m ghostnet_x1_3_imagenet_ssld
+```
+
+这样就完成了一个分类任务服务化API的部署，默认端口号为8866。
+
+**NOTE:** 如使用GPU预测，则需要在启动服务之前，请设置CUDA_VISIBLE_DEVICES环境变量，否则不用设置。
+
+## Step2: 发送预测请求
+
+配置好服务端，以下数行代码即可实现发送预测请求，获取预测结果
+
+```python
+import requests
+import json
+import cv2
+import base64
+
+import numpy as np
+
+
+def cv2_to_base64(image):
+    data = cv2.imencode('.jpg', image)[1]
+    return base64.b64encode(data.tostring()).decode('utf8')
+
+def base64_to_cv2(b64str):
+    data = base64.b64decode(b64str.encode('utf8'))
+    data = np.fromstring(data, np.uint8)
+    data = cv2.imdecode(data, cv2.IMREAD_COLOR)
+    return data
+
+# 发送HTTP请求
+org_im = cv2.imread('/PATH/TO/IMAGE')
+
+data = {'images':[cv2_to_base64(org_im)], 'top_k':2}
+headers = {"Content-type": "application/json"}
+url = "http://127.0.0.1:8866/predict/ghostnet_x1_3_imagenet_ssld"
+r = requests.post(url=url, headers=headers, data=json.dumps(data))
+data =r.json()["results"]['data']
+```
+
+### 查看代码
+
+[PaddleClas](https://github.com/PaddlePaddle/PaddleClas)
+
+### 依赖
+
+paddlepaddle >= 2.0.0
+
+paddlehub >= 2.0.0

modules/image/classification/ghostnet_x1_3_imagenet_ssld/module.py

+# copyright (c) 2021 PaddlePaddle Authors. All Rights Reserve.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#    http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import os
+import math
+from typing import Union
+
+import paddle
+import paddle.nn as nn
+import paddle.nn.functional as F
+import paddlehub.vision.transforms as T
+import numpy as np
+from paddle import ParamAttr
+from paddle.nn.initializer import Uniform, KaimingNormal
+from paddlehub.module.module import moduleinfo
+from paddlehub.module.cv_module import ImageClassifierModule
+
+
+class ConvBNLayer(nn.Layer):
+    def __init__(self, in_channels, out_channels, kernel_size, stride=1, groups=1, act="relu", name=None):
+        super(ConvBNLayer, self).__init__()
+        self._conv = nn.Conv2D(
+            in_channels=in_channels,
+            out_channels=out_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            padding=(kernel_size - 1) // 2,
+            groups=groups,
+            weight_attr=ParamAttr(initializer=KaimingNormal(), name=name + "_weights"),
+            bias_attr=False)
+        bn_name = name + "_bn"
+
+        self._batch_norm = nn.BatchNorm(
+            num_channels=out_channels,
+            act=act,
+            param_attr=ParamAttr(name=bn_name + "_scale", regularizer=paddle.regularizer.L2Decay(0.0)),
+            bias_attr=ParamAttr(name=bn_name + "_offset", regularizer=paddle.regularizer.L2Decay(0.0)),
+            moving_mean_name=bn_name + "_mean",
+            moving_variance_name=bn_name + "_variance")
+
+    def forward(self, inputs):
+        y = self._conv(inputs)
+        y = self._batch_norm(y)
+        return y
+
+
+class SEBlock(nn.Layer):
+    def __init__(self, num_channels, reduction_ratio=4, name=None):
+        super(SEBlock, self).__init__()
+        self.pool2d_gap = nn.AdaptiveAvgPool2D(1)
+        self._num_channels = num_channels
+        stdv = 1.0 / math.sqrt(num_channels * 1.0)
+        med_ch = num_channels // reduction_ratio
+        self.squeeze = nn.Linear(
+            num_channels,
+            med_ch,
+            weight_attr=ParamAttr(initializer=Uniform(-stdv, stdv), name=name + "_1_weights"),
+            bias_attr=ParamAttr(name=name + "_1_offset"))
+        stdv = 1.0 / math.sqrt(med_ch * 1.0)
+        self.excitation = nn.Linear(
+            med_ch,
+            num_channels,
+            weight_attr=ParamAttr(initializer=Uniform(-stdv, stdv), name=name + "_2_weights"),
+            bias_attr=ParamAttr(name=name + "_2_offset"))
+
+    def forward(self, inputs):
+        pool = self.pool2d_gap(inputs)
+        pool = paddle.squeeze(pool, axis=[2, 3])
+        squeeze = self.squeeze(pool)
+        squeeze = F.relu(squeeze)
+        excitation = self.excitation(squeeze)
+        excitation = paddle.clip(x=excitation, min=0, max=1)
+        excitation = paddle.unsqueeze(excitation, axis=[2, 3])
+        out = paddle.multiply(inputs, excitation)
+        return out
+
+
+class GhostModule(nn.Layer):
+    def __init__(self, in_channels, output_channels, kernel_size=1, ratio=2, dw_size=3, stride=1, relu=True, name=None):
+        super(GhostModule, self).__init__()
+        init_channels = int(math.ceil(output_channels / ratio))
+        new_channels = int(init_channels * (ratio - 1))
+        self.primary_conv = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=init_channels,
+            kernel_size=kernel_size,
+            stride=stride,
+            groups=1,
+            act="relu" if relu else None,
+            name=name + "_primary_conv")
+        self.cheap_operation = ConvBNLayer(
+            in_channels=init_channels,
+            out_channels=new_channels,
+            kernel_size=dw_size,
+            stride=1,
+            groups=init_channels,
+            act="relu" if relu else None,
+            name=name + "_cheap_operation")
+
+    def forward(self, inputs):
+        x = self.primary_conv(inputs)
+        y = self.cheap_operation(x)
+        out = paddle.concat([x, y], axis=1)
+        return out
+
+
+class GhostBottleneck(nn.Layer):
+    def __init__(self, in_channels, hidden_dim, output_channels, kernel_size, stride, use_se, name=None):
+        super(GhostBottleneck, self).__init__()
+        self._stride = stride
+        self._use_se = use_se
+        self._num_channels = in_channels
+        self._output_channels = output_channels
+        self.ghost_module_1 = GhostModule(
+            in_channels=in_channels,
+            output_channels=hidden_dim,
+            kernel_size=1,
+            stride=1,
+            relu=True,
+            name=name + "_ghost_module_1")
+        if stride == 2:
+            self.depthwise_conv = ConvBNLayer(
+                in_channels=hidden_dim,
+                out_channels=hidden_dim,
+                kernel_size=kernel_size,
+                stride=stride,
+                groups=hidden_dim,
+                act=None,
+                name=name + "_depthwise_depthwise"  # looks strange due to an old typo, will be fixed later.
+            )
+        if use_se:
+            self.se_block = SEBlock(num_channels=hidden_dim, name=name + "_se")
+        self.ghost_module_2 = GhostModule(
+            in_channels=hidden_dim,
+            output_channels=output_channels,
+            kernel_size=1,
+            relu=False,
+            name=name + "_ghost_module_2")
+        if stride != 1 or in_channels != output_channels:
+            self.shortcut_depthwise = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=in_channels,
+                kernel_size=kernel_size,
+                stride=stride,
+                groups=in_channels,
+                act=None,
+                name=name + "_shortcut_depthwise_depthwise"  # looks strange due to an old typo, will be fixed later.
+            )
+            self.shortcut_conv = ConvBNLayer(
+                in_channels=in_channels,
+                out_channels=output_channels,
+                kernel_size=1,
+                stride=1,
+                groups=1,
+                act=None,
+                name=name + "_shortcut_conv")
+
+    def forward(self, inputs):
+        x = self.ghost_module_1(inputs)
+        if self._stride == 2:
+            x = self.depthwise_conv(x)
+        if self._use_se:
+            x = self.se_block(x)
+        x = self.ghost_module_2(x)
+        if self._stride == 1 and self._num_channels == self._output_channels:
+            shortcut = inputs
+        else:
+            shortcut = self.shortcut_depthwise(inputs)
+            shortcut = self.shortcut_conv(shortcut)
+        return paddle.add(x=x, y=shortcut)
+
+
+@moduleinfo(
+    name="ghostnet_x1_3_imagenet_ssld",
+    type="CV/classification",
+    author="paddlepaddle",
+    author_email="",
+    summary="ghostnet_x1_3_imagenet_ssld is a classification model, "
+    "this module is trained with Imagenet dataset.",
+    version="1.0.0",
+    meta=ImageClassifierModule)
+class GhostNet(nn.Layer):
+    def __init__(self, label_list: list = None, load_checkpoint: str = None):
+        super(GhostNet, self).__init__()
+
+        if label_list is not None:
+            self.labels = label_list
+            class_dim = len(self.labels)
+        else:
+            label_list = []
+            label_file = os.path.join(self.directory, 'label_list.txt')
+            files = open(label_file)
+            for line in files.readlines():
+                line = line.strip('\n')
+                label_list.append(line)
+            self.labels = label_list
+            class_dim = len(self.labels)
+
+        self.cfgs = [
+            # k, t, c, SE, s
+            [3, 16, 16, 0, 1],
+            [3, 48, 24, 0, 2],
+            [3, 72, 24, 0, 1],
+            [5, 72, 40, 1, 2],
+            [5, 120, 40, 1, 1],
+            [3, 240, 80, 0, 2],
+            [3, 200, 80, 0, 1],
+            [3, 184, 80, 0, 1],
+            [3, 184, 80, 0, 1],
+            [3, 480, 112, 1, 1],
+            [3, 672, 112, 1, 1],
+            [5, 672, 160, 1, 2],
+            [5, 960, 160, 0, 1],
+            [5, 960, 160, 1, 1],
+            [5, 960, 160, 0, 1],
+            [5, 960, 160, 1, 1]
+        ]
+        self.scale = 1.3
+        output_channels = int(self._make_divisible(16 * self.scale, 4))
+        self.conv1 = ConvBNLayer(
+            in_channels=3, out_channels=output_channels, kernel_size=3, stride=2, groups=1, act="relu", name="conv1")
+        # build inverted residual blocks
+        idx = 0
+        self.ghost_bottleneck_list = []
+        for k, exp_size, c, use_se, s in self.cfgs:
+            in_channels = output_channels
+            output_channels = int(self._make_divisible(c * self.scale, 4))
+            hidden_dim = int(self._make_divisible(exp_size * self.scale, 4))
+            ghost_bottleneck = self.add_sublayer(
+                name="_ghostbottleneck_" + str(idx),
+                sublayer=GhostBottleneck(
+                    in_channels=in_channels,
+                    hidden_dim=hidden_dim,
+                    output_channels=output_channels,
+                    kernel_size=k,
+                    stride=s,
+                    use_se=use_se,
+                    name="_ghostbottleneck_" + str(idx)))
+            self.ghost_bottleneck_list.append(ghost_bottleneck)
+            idx += 1
+        # build last several layers
+        in_channels = output_channels
+        output_channels = int(self._make_divisible(exp_size * self.scale, 4))
+        self.conv_last = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=output_channels,
+            kernel_size=1,
+            stride=1,
+            groups=1,
+            act="relu",
+            name="conv_last")
+        self.pool2d_gap = nn.AdaptiveAvgPool2D(1)
+        in_channels = output_channels
+        self._fc0_output_channels = 1280
+        self.fc_0 = ConvBNLayer(
+            in_channels=in_channels,
+            out_channels=self._fc0_output_channels,
+            kernel_size=1,
+            stride=1,
+            act="relu",
+            name="fc_0")
+        self.dropout = nn.Dropout(p=0.2)
+        stdv = 1.0 / math.sqrt(self._fc0_output_channels * 1.0)
+        self.fc_1 = nn.Linear(
+            self._fc0_output_channels,
+            class_dim,
+            weight_attr=ParamAttr(name="fc_1_weights", initializer=Uniform(-stdv, stdv)),
+            bias_attr=ParamAttr(name="fc_1_offset"))
+
+        if load_checkpoint is not None:
+            self.model_dict = paddle.load(load_checkpoint)
+            self.set_dict(self.model_dict)
+            print("load custom checkpoint success")
+        else:
+            checkpoint = os.path.join(self.directory, 'model.pdparams')
+            self.model_dict = paddle.load(checkpoint)
+            self.set_dict(self.model_dict)
+            print("load pretrained checkpoint success")
+
+    def transforms(self, images: Union[str, np.ndarray]):
+        transforms = T.Compose([
+            T.Resize((256, 256)),
+            T.CenterCrop(224),
+            T.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
+        ],
+                               to_rgb=True)
+        return transforms(images).astype('float32')
+
+    def forward(self, inputs):
+        x = self.conv1(inputs)
+        for ghost_bottleneck in self.ghost_bottleneck_list:
+            x = ghost_bottleneck(x)
+        x = self.conv_last(x)
+        feature = self.pool2d_gap(x)
+        x = self.fc_0(feature)
+        x = self.dropout(x)
+        x = paddle.reshape(x, shape=[-1, self._fc0_output_channels])
+        x = self.fc_1(x)
+        return x, feature
+
+    def _make_divisible(self, v, divisor, min_value=None):
+        """
+        This function is taken from the original tf repo.
+        It ensures that all layers have a channel number that is divisible by 8
+        It can be seen here:
+        https://github.com/tensorflow/models/blob/master/research/slim/nets/mobilenet/mobilenet.py
+        """
+        if min_value is None:
+            min_value = divisor
+        new_v = max(min_value, int(v + divisor / 2) // divisor * divisor)
+        # Make sure that round down does not go down by more than 10%.
+        if new_v < 0.9 * v:
+            new_v += divisor
+        return new_v

modules/image/semantic_segmentation/deeplabv3p_resnet50_voc/layers.py

@@ -15,7 +15,6 @@
 import paddle
 import paddle.nn as nn
 import paddle.nn.functional as F
-from paddle.nn.layer import activation
 from paddle.nn import Conv2D, AvgPool2D
 
 
@@ -214,7 +213,7 @@ class Activation(nn.Layer):
         super(Activation, self).__init__()
 
         self._act = act
-        upper_act_names = activation.__all__
+        upper_act_names = nn.layer.activation.__dict__.keys()
         lower_act_names = [act.lower() for act in upper_act_names]
         act_dict = dict(zip(lower_act_names, upper_act_names))
 

modules/image/text_recognition/chinese_ocr_db_crnn_server/utils.py

@@ -172,6 +172,6 @@ def sorted_boxes(dt_boxes):
 
 def base64_to_cv2(b64str):
     data = base64.b64decode(b64str.encode('utf8'))
-    data = np.fromstring(data, np.uint8)
+    data = np.frombuffer(data, np.uint8)
     data = cv2.imdecode(data, cv2.IMREAD_COLOR)
     return data

modules/text/machine_translation/transformer/en-de/requirements.txt

+sacremoses
+subword-nmt