Merge pull request #1687 from rainyfly/add_EnlightenGAN_module

add EnlightenGAN module

Merge pull request #1687 from rainyfly/add_EnlightenGAN_module
add EnlightenGAN module
2c609e01 · KP · GitHub · 27b1e6d9 · 658ec814 · 2c609e01
4 changed file
--- a/modules/image/image_processing/enlightengan/README.md
+++ b/modules/image/image_processing/enlightengan/README.md
+# enlightengan
+
+|模型名称|enlightengan|
+| :--- | :---: |
+|类别|图像 - 暗光增强|
+|网络|EnlightenGAN|
+|数据集|-|
+|是否支持Fine-tuning|否|
+|模型大小|83MB|
+|最新更新日期|2021-11-04|
+|数据指标|-|
+
+
+## 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。
+
+  - 更多详情参考：[EnlightenGAN: Deep Light Enhancement without Paired Supervision](https://arxiv.org/abs/1906.06972)
+
+
+
+## 二、安装
+
+- ### 1、环境依赖  
+  - onnxruntime
+  - x2paddle
+  - pillow
+
+- ### 2、安装
+
+  - ```shell
+    $ 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)
+
+## 三、模型API预测
+
+- ### 1、命令行预测
+
+  - ```shell
+    # Read from a file
+    $ 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)
+
+- ### 2、预测代码示例
+
+  - ```python
+    import paddlehub as hub
+
+    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(images=None, paths=None, output_dir='./enlightening_result/', use_gpu=False, visualization=True)
+    ```
+    - 暗光增强API。
+
+    - **参数**
+
+      - 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"])
+    ```
+
+## 五、更新历史
+
+* 1.0.0
+
+  初始发布
+
+  - ```shell
+    $ hub install enlightengan==1.0.0
+    ```
--- a/modules/image/image_processing/enlightengan/enlighten_inference/pd_model/x2paddle_code.py
+++ b/modules/image/image_processing/enlightengan/enlighten_inference/pd_model/x2paddle_code.py
+import paddle
+import math
+
+
+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
--- a/modules/image/image_processing/enlightengan/module.py
+++ b/modules/image/image_processing/enlightengan/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 argparse
+import os
+
+import cv2
+import numpy as np
+import paddle
+
+import paddlehub as hub
+from .enlighten_inference.pd_model.x2paddle_code import ONNXModel
+from .util import base64_to_cv2
+from paddlehub.module.module import moduleinfo
+from paddlehub.module.module import runnable
+from paddlehub.module.module import serving
+
+
+@moduleinfo(name="enlightengan",
+            type="CV/enlighten",
+            author="paddlepaddle",
+            author_email="",
+            summary="",
+            version="1.0.0")
+class EnlightenGAN:
+
+    def __init__(self):
+        self.pretrained_model = os.path.join(self.directory, "enlighten_inference/pd_model")
+        self.model = ONNXModel()
+        params = paddle.load(os.path.join(self.pretrained_model, 'model.pdparams'))
+        self.model.set_dict(params, use_structured_name=True)
+
+    def enlightening(self,
+                     images: list = None,
+                     paths: list = None,
+                     output_dir: str = './enlightening_result/',
+                     use_gpu: bool = False,
+                     visualization: bool = True):
+        '''
+        enlighten images in the low-light scene.
+
+        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 (str): the dir to save the results
+        use_gpu (bool): if True, use gpu to perform the computation, otherwise cpu.
+        visualization (bool): if True, save results in output_dir.
+        '''
+        results = []
+        paddle.disable_static()
+        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.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 = cv2.imread(path)[:, :, ::-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 visualization == True:
+            if not os.path.exists(output_dir):
+                os.makedirs(output_dir, exist_ok=True)
+            for i, out in enumerate(results):
+                cv2.imwrite(os.path.join(output_dir, 'output_{}.png'.format(i)), out[:, :, ::-1])
+
+        return results
+
+    @runnable
+    def run_cmd(self, argvs: list):
+        """
+        Run as a command.
+        """
+        self.parser = argparse.ArgumentParser(description="Run the {} module.".format(self.name),
+                                              prog='hub run {}'.format(self.name),
+                                              usage='%(prog)s',
+                                              add_help=True)
+
+        self.arg_input_group = self.parser.add_argument_group(title="Input options", description="Input data. Required")
+        self.arg_config_group = self.parser.add_argument_group(
+            title="Config options", description="Run configuration for controlling module behavior, not required.")
+        self.add_module_config_arg()
+        self.add_module_input_arg()
+        self.args = self.parser.parse_args(argvs)
+        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):
+        """
+        Add the command config options.
+        """
+        self.arg_config_group.add_argument('--use_gpu', action='store_true', help="use GPU or not")
+
+        self.arg_config_group.add_argument('--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):
+        """
+        Add the command input options.
+        """
+        self.arg_input_group.add_argument('--input_path', type=str, help="path to input image.")
--- a/modules/image/image_processing/enlightengan/util.py
+++ b/modules/image/image_processing/enlightengan/util.py
+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