Skip to content

P
PaddleHub

PaddlePaddle / PaddleHub
大约 1 年 前同步成功

通知 282
Star 12117
Fork 2091
P
PaddleHub
提交 faea8791 编写于 作者: C chenjian
浏览文件
操作

update according to review

上级 87ea363f
隐藏空白更改
内联 并排
Showing with 288 addition and 236 deletion
modules/image/image_processing/EnlightenGAN/enlighten_inference/pd_model/x2paddle_code.py 已删除 100755 → 0
浏览文件 @ 87ea363f
import paddle
import math
from x2paddle.op_mapper.onnx2paddle import onnx_custom_layer as x2paddle_nn
class ONNXModel(paddle.nn.Layer):
def __init__(self):
super(ONNXModel, self).__init__()
self.conv0 = paddle.nn.Conv2D(in_channels=3, out_channels=3, kernel_size=[1, 1], groups=3)
self.pool0 = paddle.nn.MaxPool2D(kernel_size=[2, 2], stride=2)
self.pool1 = paddle.nn.MaxPool2D(kernel_size=[2, 2], stride=2)
self.conv1 = paddle.nn.Conv2D(in_channels=4, out_channels=32, kernel_size=[3, 3], padding=1)
self.pool2 = paddle.nn.MaxPool2D(kernel_size=[2, 2], stride=2)
self.leakyrelu0 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.pool3 = paddle.nn.MaxPool2D(kernel_size=[2, 2], stride=2)
self.batchnorm0 = paddle.nn.BatchNorm(
num_channels=32, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv2 = paddle.nn.Conv2D(in_channels=32, out_channels=32, kernel_size=[3, 3], padding=1)
self.leakyrelu1 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm1 = paddle.nn.BatchNorm(
num_channels=32, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.pool4 = paddle.nn.MaxPool2D(kernel_size=[2, 2], stride=2)
self.conv3 = paddle.nn.Conv2D(in_channels=32, out_channels=64, kernel_size=[3, 3], padding=1)
self.leakyrelu2 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm2 = paddle.nn.BatchNorm(
num_channels=64, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv4 = paddle.nn.Conv2D(in_channels=64, out_channels=64, kernel_size=[3, 3], padding=1)
self.leakyrelu3 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm3 = paddle.nn.BatchNorm(
num_channels=64, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.pool5 = paddle.nn.MaxPool2D(kernel_size=[2, 2], stride=2)
self.conv5 = paddle.nn.Conv2D(in_channels=64, out_channels=128, kernel_size=[3, 3], padding=1)
self.leakyrelu4 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm4 = paddle.nn.BatchNorm(
num_channels=128, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv6 = paddle.nn.Conv2D(in_channels=128, out_channels=128, kernel_size=[3, 3], padding=1)
self.leakyrelu5 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm5 = paddle.nn.BatchNorm(
num_channels=128, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.pool6 = paddle.nn.MaxPool2D(kernel_size=[2, 2], stride=2)
self.conv7 = paddle.nn.Conv2D(in_channels=128, out_channels=256, kernel_size=[3, 3], padding=1)
self.leakyrelu6 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm6 = paddle.nn.BatchNorm(
num_channels=256, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv8 = paddle.nn.Conv2D(in_channels=256, out_channels=256, kernel_size=[3, 3], padding=1)
self.leakyrelu7 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm7 = paddle.nn.BatchNorm(
num_channels=256, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.pool7 = paddle.nn.MaxPool2D(kernel_size=[2, 2], stride=2)
self.conv9 = paddle.nn.Conv2D(in_channels=256, out_channels=512, kernel_size=[3, 3], padding=1)
self.leakyrelu8 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm8 = paddle.nn.BatchNorm(
num_channels=512, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv10 = paddle.nn.Conv2D(in_channels=512, out_channels=512, kernel_size=[3, 3], padding=1)
self.leakyrelu9 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm9 = paddle.nn.BatchNorm(
num_channels=512, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv11 = paddle.nn.Conv2D(in_channels=512, out_channels=256, kernel_size=[3, 3], padding=1)
self.conv12 = paddle.nn.Conv2D(in_channels=512, out_channels=256, kernel_size=[3, 3], padding=1)
self.leakyrelu10 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm10 = paddle.nn.BatchNorm(
num_channels=256, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv13 = paddle.nn.Conv2D(in_channels=256, out_channels=256, kernel_size=[3, 3], padding=1)
self.leakyrelu11 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm11 = paddle.nn.BatchNorm(
num_channels=256, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv14 = paddle.nn.Conv2D(in_channels=256, out_channels=128, kernel_size=[3, 3], padding=1)
self.conv15 = paddle.nn.Conv2D(in_channels=256, out_channels=128, kernel_size=[3, 3], padding=1)
self.leakyrelu12 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm12 = paddle.nn.BatchNorm(
num_channels=128, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv16 = paddle.nn.Conv2D(in_channels=128, out_channels=128, kernel_size=[3, 3], padding=1)
self.leakyrelu13 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm13 = paddle.nn.BatchNorm(
num_channels=128, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv17 = paddle.nn.Conv2D(in_channels=128, out_channels=64, kernel_size=[3, 3], padding=1)
self.conv18 = paddle.nn.Conv2D(in_channels=128, out_channels=64, kernel_size=[3, 3], padding=1)
self.leakyrelu14 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm14 = paddle.nn.BatchNorm(
num_channels=64, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv19 = paddle.nn.Conv2D(in_channels=64, out_channels=64, kernel_size=[3, 3], padding=1)
self.leakyrelu15 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm15 = paddle.nn.BatchNorm(
num_channels=64, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv20 = paddle.nn.Conv2D(in_channels=64, out_channels=32, kernel_size=[3, 3], padding=1)
self.conv21 = paddle.nn.Conv2D(in_channels=64, out_channels=32, kernel_size=[3, 3], padding=1)
self.leakyrelu16 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.batchnorm16 = paddle.nn.BatchNorm(
num_channels=32, momentum=0.8999999761581421, epsilon=9.999999747378752e-06, is_test=True)
self.conv22 = paddle.nn.Conv2D(in_channels=32, out_channels=32, kernel_size=[3, 3], padding=1)
self.leakyrelu17 = paddle.nn.LeakyReLU(negative_slope=0.20000000298023224)
self.conv23 = paddle.nn.Conv2D(in_channels=32, out_channels=3, kernel_size=[1, 1])
def forward(self, x2paddle_input):
x2paddle_137 = paddle.full(dtype='float32', shape=[1], fill_value=1.0)
x2paddle_145 = paddle.full(dtype='float32', shape=[1], fill_value=0.29899999499320984)
x2paddle_147 = paddle.full(dtype='float32', shape=[1], fill_value=0.5870000123977661)
x2paddle_150 = paddle.full(dtype='float32', shape=[1], fill_value=0.11400000005960464)
x2paddle_153 = paddle.full(dtype='float32', shape=[1], fill_value=2.0)
x2paddle_155 = paddle.full(dtype='float32', shape=[1], fill_value=1.0)
x2paddle_256 = paddle.full(dtype='float32', shape=[1], fill_value=1.0)
x2paddle_134 = self.conv0(x2paddle_input)
x2paddle_135, = paddle.split(x=x2paddle_134, num_or_sections=[1])
x2paddle_257 = paddle.multiply(x=x2paddle_134, y=x2paddle_256)
x2paddle_136 = paddle.squeeze(x=x2paddle_135, axis=[0])
x2paddle_138 = paddle.add(x=x2paddle_136, y=x2paddle_137)
x2paddle_139_p0, x2paddle_139_p1, x2paddle_139_p2 = paddle.split(x=x2paddle_138, num_or_sections=[1, 1, 1])
x2paddle_142 = paddle.squeeze(x=x2paddle_139_p0, axis=[0])
x2paddle_143 = paddle.squeeze(x=x2paddle_139_p1, axis=[0])
x2paddle_144 = paddle.squeeze(x=x2paddle_139_p2, axis=[0])
x2paddle_146 = paddle.multiply(x=x2paddle_142, y=x2paddle_145)
x2paddle_148 = paddle.multiply(x=x2paddle_143, y=x2paddle_147)
x2paddle_151 = paddle.multiply(x=x2paddle_144, y=x2paddle_150)
x2paddle_149 = paddle.add(x=x2paddle_146, y=x2paddle_148)
x2paddle_152 = paddle.add(x=x2paddle_149, y=x2paddle_151)
x2paddle_154 = paddle.divide(x=x2paddle_152, y=x2paddle_153)
x2paddle_156 = paddle.subtract(x=x2paddle_155, y=x2paddle_154)
x2paddle_157 = paddle.unsqueeze(x=x2paddle_156, axis=[0])
x2paddle_158 = paddle.unsqueeze(x=x2paddle_157, axis=[0])
x2paddle_159 = self.pool0(x2paddle_158)
x2paddle_163 = paddle.concat(x=[x2paddle_134, x2paddle_158], axis=1)
x2paddle_160 = self.pool1(x2paddle_159)
x2paddle_164 = self.conv1(x2paddle_163)
x2paddle_161 = self.pool2(x2paddle_160)
x2paddle_165 = self.leakyrelu0(x2paddle_164)
x2paddle_162 = self.pool3(x2paddle_161)
x2paddle_166 = self.batchnorm0(x2paddle_165)
x2paddle_167 = self.conv2(x2paddle_166)
x2paddle_168 = self.leakyrelu1(x2paddle_167)
x2paddle_169 = self.batchnorm1(x2paddle_168)
x2paddle_170 = self.pool4(x2paddle_169)
x2paddle_246 = paddle.multiply(x=x2paddle_169, y=x2paddle_158)
x2paddle_171 = self.conv3(x2paddle_170)
x2paddle_172 = self.leakyrelu2(x2paddle_171)
x2paddle_173 = self.batchnorm2(x2paddle_172)
x2paddle_174 = self.conv4(x2paddle_173)
x2paddle_175 = self.leakyrelu3(x2paddle_174)
x2paddle_176 = self.batchnorm3(x2paddle_175)
x2paddle_177 = self.pool5(x2paddle_176)
x2paddle_232 = paddle.multiply(x=x2paddle_176, y=x2paddle_159)
x2paddle_178 = self.conv5(x2paddle_177)
x2paddle_179 = self.leakyrelu4(x2paddle_178)
x2paddle_180 = self.batchnorm4(x2paddle_179)
x2paddle_181 = self.conv6(x2paddle_180)
x2paddle_182 = self.leakyrelu5(x2paddle_181)
x2paddle_183 = self.batchnorm5(x2paddle_182)
x2paddle_184 = self.pool6(x2paddle_183)
x2paddle_218 = paddle.multiply(x=x2paddle_183, y=x2paddle_160)
x2paddle_185 = self.conv7(x2paddle_184)
x2paddle_186 = self.leakyrelu6(x2paddle_185)
x2paddle_187 = self.batchnorm6(x2paddle_186)
x2paddle_188 = self.conv8(x2paddle_187)
x2paddle_189 = self.leakyrelu7(x2paddle_188)
x2paddle_190 = self.batchnorm7(x2paddle_189)
x2paddle_191 = self.pool7(x2paddle_190)
x2paddle_204 = paddle.multiply(x=x2paddle_190, y=x2paddle_161)
x2paddle_192 = self.conv9(x2paddle_191)
x2paddle_193 = self.leakyrelu8(x2paddle_192)
x2paddle_194 = self.batchnorm8(x2paddle_193)
x2paddle_195 = paddle.multiply(x=x2paddle_194, y=x2paddle_162)
x2paddle_196 = self.conv10(x2paddle_195)
x2paddle_197 = self.leakyrelu9(x2paddle_196)
x2paddle_198 = self.batchnorm9(x2paddle_197)
x2paddle_203 = paddle.nn.functional.interpolate(x=x2paddle_198, scale_factor=[2.0, 2.0], mode='bilinear')
x2paddle_205 = self.conv11(x2paddle_203)
x2paddle_206 = paddle.concat(x=[x2paddle_205, x2paddle_204], axis=1)
x2paddle_207 = self.conv12(x2paddle_206)
x2paddle_208 = self.leakyrelu10(x2paddle_207)
x2paddle_209 = self.batchnorm10(x2paddle_208)
x2paddle_210 = self.conv13(x2paddle_209)
x2paddle_211 = self.leakyrelu11(x2paddle_210)
x2paddle_212 = self.batchnorm11(x2paddle_211)
x2paddle_217 = paddle.nn.functional.interpolate(x=x2paddle_212, scale_factor=[2.0, 2.0], mode='bilinear')
x2paddle_219 = self.conv14(x2paddle_217)
x2paddle_220 = paddle.concat(x=[x2paddle_219, x2paddle_218], axis=1)
x2paddle_221 = self.conv15(x2paddle_220)
x2paddle_222 = self.leakyrelu12(x2paddle_221)
x2paddle_223 = self.batchnorm12(x2paddle_222)
x2paddle_224 = self.conv16(x2paddle_223)
x2paddle_225 = self.leakyrelu13(x2paddle_224)
x2paddle_226 = self.batchnorm13(x2paddle_225)
x2paddle_231 = paddle.nn.functional.interpolate(x=x2paddle_226, scale_factor=[2.0, 2.0], mode='bilinear')
x2paddle_233 = self.conv17(x2paddle_231)
x2paddle_234 = paddle.concat(x=[x2paddle_233, x2paddle_232], axis=1)
x2paddle_235 = self.conv18(x2paddle_234)
x2paddle_236 = self.leakyrelu14(x2paddle_235)
x2paddle_237 = self.batchnorm14(x2paddle_236)
x2paddle_238 = self.conv19(x2paddle_237)
x2paddle_239 = self.leakyrelu15(x2paddle_238)
x2paddle_240 = self.batchnorm15(x2paddle_239)
x2paddle_245 = paddle.nn.functional.interpolate(x=x2paddle_240, scale_factor=[2.0, 2.0], mode='bilinear')
x2paddle_247 = self.conv20(x2paddle_245)
x2paddle_248 = paddle.concat(x=[x2paddle_247, x2paddle_246], axis=1)
x2paddle_249 = self.conv21(x2paddle_248)
x2paddle_250 = self.leakyrelu16(x2paddle_249)
x2paddle_251 = self.batchnorm16(x2paddle_250)
x2paddle_252 = self.conv22(x2paddle_251)
x2paddle_253 = self.leakyrelu17(x2paddle_252)
x2paddle_254 = self.conv23(x2paddle_253)
x2paddle_255 = paddle.multiply(x=x2paddle_254, y=x2paddle_158)
x2paddle_output = paddle.add(x=x2paddle_255, y=x2paddle_257)
return x2paddle_output, x2paddle_255
def main(x2paddle_input):
# There are 1 inputs.
# x2paddle_input: shape-[-1, 3, 512, 512], type-float32.
paddle.disable_static()
params = paddle.load('/work/ToTransferInHub/EnlightenGAN-inference/enlighten_inference/pd_model/model.pdparams')
model = ONNXModel()
model.set_dict(params, use_structured_name=True)
model.eval()
out = model(x2paddle_input)
return out
if __name__ == '__main__':
inputtensor = paddle.randn([1, 3, 512, 512])
main(inputtensor)
modules/image/image_processing/EnlightenGAN/README.md → modules/image/image_processing/enlightengan/README.md
浏览文件 @ faea8791
# EnlightenGAN
# enlightengan
|模型名称|EnlightenGAN|
|模型名称|enlightengan|
| :--- | :---: |
|类别|图像 - 暗光增强|
|网络|EnlightenGAN|
|数据集||
|数据集|-|
|是否支持Fine-tuning|否|
|模型大小|83MB|
|最新更新日期|2021-11-04|
......@@ -13,6 +13,19 @@
## 一、模型基本信息
- ### 应用效果展示
- 样例结果示例：
<p align="center">
<img src="https://user-images.githubusercontent.com/22424850/142827116-76d713c6-94d9-410d-830a-65135cd856b8.jpeg" width = "450" height = "300" hspace='10'/>
<br />
输入图像
<br />
<img src="https://user-images.githubusercontent.com/22424850/142827262-97317323-f6bd-4aa4-b7ac-c69436c4d576.png" width = "450" height = "300" hspace='10'/>
<br />
输出图像
<br />
</p>
- ### 模型介绍
- EnlightenGAN使用非成对的数据进行训练，通过设计自特征保留损失函数和自约束注意力机制，训练的网络可以应用到多种场景下的暗光增强中。
......@@ -31,7 +44,7 @@
- ### 2、安装
- ```shell
$ hub install EnlightenGAN
$ hub install enlightengan
```
- 如您安装时遇到问题，可参考：[零基础windows安装](../../../../docs/docs_ch/get_start/windows_quickstart.md)
| [零基础Linux安装](../../../../docs/docs_ch/get_start/linux_quickstart.md) | [零基础MacOS安装](../../../../docs/docs_ch/get_start/mac_quickstart.md)
......@@ -42,7 +55,7 @@
- ```shell
# Read from a file
$ hub run EnlightenGAN --input_path "/PATH/TO/IMAGE"
$ hub run enlightengan --input_path "/PATH/TO/IMAGE"
```
- 通过命令行方式实现暗光增强模型的调用，更多请见 [PaddleHub命令行指令](../../../../docs/docs_ch/tutorial/cmd_usage.rst)
......@@ -51,34 +64,73 @@
- ```python
import paddlehub as hub
enlightener = hub.Module(name="EnlightenGAN")
input_path = "/PATH/TO/IMAGE"
# Read from a raw file
enlightener.enlightening(input_path, output_path='./enlightening_result.png', use_gpu=True)
enlightener = hub.Module(name="enlightengan")
input_path = ["/PATH/TO/IMAGE"]
# Read from a file
enlightener.enlightening(paths=input_path, output_dir='./enlightening_result/', use_gpu=True)
```
- ### 3、API
- ```python
def enlightening(input_path, output_path='./enlightening_result.png', use_gpu=False)
def enlightening(images=None, paths=None, output_dir='./enlightening_result/', use_gpu=False, visualization=True)
```
- 暗光增强API。
- **参数**
- input\_path (str): 输入图像文件的路径; <br/>
- output\_path (str): 结果保存的路径, 需要指定输出文件名； <br/>
- images (list\[numpy.ndarray\]): 图片数据，ndarray.shape 为 \[H, W, C\]；<br/>
- paths (list\[str\]): 图片的路径；<br/>
- output\_dir (str): 结果保存的路径； <br/>
- use\_gpu (bool): 是否使用 GPU；<br/>
- visualization(bool): 是否保存结果到本地文件夹
## 四、服务部署
- PaddleHub Serving可以部署一个在线图像风格转换服务。
- ### 第一步：启动PaddleHub Serving
- 运行启动命令：
- ```shell
$ hub serving start -m enlightengan
```
- 这样就完成了一个图像风格转换的在线服务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。
- ### 第二步：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
def cv2_to_base64(image):
data = cv2.imencode('.jpg', image)[1]
return base64.b64encode(data.tostring()).decode('utf8')
# 发送HTTP请求
data = {'images':[cv2_to_base64(cv2.imread("/PATH/TO/IMAGE"))]}
headers = {"Content-type": "application/json"}
url = "http://127.0.0.1:8866/predict/enlightengan"
r = requests.post(url=url, headers=headers, data=json.dumps(data))
# 打印预测结果
print(r.json()["results"])
## 四、更新历史
## 五、更新历史
* 1.0.0
初始发布
- ```shell
$ hub install EnlightenGAN==1.0.0
$ hub install enlightengan==1.0.0
```
modules/image/image_processing/EnlightenGAN/x2paddle_code.py → modules/image/image_processing/enlightengan/enlighten_inference/pd_model/x2paddle_code.py
浏览文件 @ faea8791
......@@ -200,20 +200,3 @@ class ONNXModel(paddle.nn.Layer):
x2paddle_255 = paddle.multiply(x=x2paddle_254, y=x2paddle_158)
x2paddle_output = paddle.add(x=x2paddle_255, y=x2paddle_257)
return x2paddle_output, x2paddle_255
def main(x2paddle_input):
# There are 1 inputs.
# x2paddle_input: shape-[-1, 3, 512, 512], type-float32.
paddle.disable_static()
params = paddle.load('/work/ToTransferInHub/EnlightenGAN-inference/enlighten_inference/pd_model/model.pdparams')
model = ONNXModel()
model.set_dict(params, use_structured_name=True)
model.eval()
out = model(x2paddle_input)
return out
if __name__ == '__main__':
inputtensor = paddle.randn([1, 3, 512, 512])
main(inputtensor)
modules/image/image_processing/EnlightenGAN/module.py → modules/image/image_processing/enlightengan/module.py
浏览文件 @ faea8791
......@@ -17,11 +17,13 @@ import argparse
import paddle
import paddlehub as hub
from paddlehub.module.module import moduleinfo, runnable
from PIL import Image
from paddlehub.module.module import moduleinfo, runnable, serving
import numpy as np
from PIL import Image
from .enlighten_inference import EnlightenOnnxModel
from .enlighten_inference.pd_model.x2paddle_code import ONNXModel
from .util import base64_to_cv2
@moduleinfo(
......@@ -30,36 +32,64 @@ class EnlightenGAN:
def __init__(self):
self.pretrained_model = os.path.join(self.directory, "enlighten_inference/pd_model")
def enlightening(self, input_path, output_path='./enlightening_result.png', use_gpu=False):
def enlightening(self,
images=None,
paths=None,
output_dir='./enlightening_result/',
use_gpu=False,
visualization=True):
'''
enlighten a image in the low-light scene.
enlighten images in the low-light scene.
input_path: the image path
output_path: the path to save the results
images (list[numpy.ndarray]): data of images, shape of each is [H, W, C], color space must be BGR(read by cv2).
paths (list[str]): paths to images
output_dir: the dir to save the results
use_gpu: if True, use gpu to perform the computation, otherwise cpu.
visualization: if True, save results in output_dir.
'''
results = []
paddle.disable_static()
img = np.array(Image.open(input_path))
img = np.expand_dims(np.transpose(img, (2, 0, 1)).astype(np.float32) / 255., 0)
inputtensor = paddle.to_tensor(img)
place = 'gpu:0' if use_gpu else 'cpu'
place = paddle.set_device(place)
if images == None and paths == None:
print('No image provided. Please input an image or a image path.')
return
self.model = ONNXModel()
params = paddle.load(os.path.join(self.pretrained_model, 'model.pdparams'))
model = ONNXModel()
model.set_dict(params, use_structured_name=True)
model.eval()
out, out1 = model(inputtensor)
out = out.numpy()[0]
out = (np.transpose(out, (1, 2, 0)) + 1) / 2.0 * 255.0
out = np.clip(out, 0, 255)
out = out.astype('uint8')
print('enlighten Over.')
try:
Image.fromarray(out).save(os.path.join(output_path))
print('Image saved in {}'.format(output_path))
except:
print('Save image failed. Please check the output_path, should\
be image format ext, e.g. png. current output path {}'.format(output_path))
return out
self.model.set_dict(params, use_structured_name=True)
self.model.eval()
if images != None:
for image in images:
image = image[:, :, ::-1]
image = np.expand_dims(np.transpose(image, (2, 0, 1)).astype(np.float32) / 255., 0)
inputtensor = paddle.to_tensor(image)
out, out1 = self.model(inputtensor)
out = out.numpy()[0]
out = (np.transpose(out, (1, 2, 0)) + 1) / 2.0 * 255.0
out = np.clip(out, 0, 255)
out = out.astype('uint8')
results.append(out)
if paths != None:
for path in paths:
image = np.array(Image.open(path))
image = np.expand_dims(np.transpose(image, (2, 0, 1)).astype(np.float32) / 255., 0)
inputtensor = paddle.to_tensor(image)
out, out1 = self.model(inputtensor)
out = out.numpy()[0]
out = (np.transpose(out, (1, 2, 0)) + 1) / 2.0 * 255.0
out = np.clip(out, 0, 255)
out = out.astype('uint8')
results.append(out)
if visualization == True:
if not os.path.exists(output_dir):
os.makedirs(output_dir, exist_ok=True)
for i, out in enumerate(results):
Image.fromarray(out).save(os.path.join(os.path.join(output_dir, 'output_{}.png'.format(i))))
return results
@runnable
def run_cmd(self, argvs: list):
......@@ -78,7 +108,22 @@ class EnlightenGAN:
self.add_module_config_arg()
self.add_module_input_arg()
self.args = self.parser.parse_args(argvs)
self.enlightening(input_path=self.args.input_path, output_path=self.args.output_path, use_gpu=self.args.use_gpu)
results = self.enlightening(
paths=[self.args.input_path],
output_dir=self.args.output_dir,
use_gpu=self.args.use_gpu,
visualization=self.args.visualization)
return results
@serving
def serving_method(self, images, **kwargs):
"""
Run as a service.
"""
images_decode = [base64_to_cv2(image) for image in images]
results = self.enlightening(images=images_decode, **kwargs)
tolist = [result.tolist() for result in results]
return tolist
def add_module_config_arg(self):
"""
......@@ -87,7 +132,8 @@ class EnlightenGAN:
self.arg_config_group.add_argument('--use_gpu', action='store_true', help="use GPU or not")
self.arg_config_group.add_argument(
'--output_path', type=str, default='enlightening_result.png', help='output path for saving result.')
'--output_dir', type=str, default='enlightening_result', help='output directory for saving result.')
self.arg_config_group.add_argument('--visualization', type=bool, default=False, help='save results or not.')
def add_module_input_arg(self):
"""
......
modules/image/image_processing/enlightengan/util.py 0 → 100644
浏览文件 @ faea8791
import base64
import cv2
import numpy as np
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
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或