[Cherry pick] ac readme and seg readme (#1280)

* optimize segmentic readme (#1273)

* optimize act readme (#1265)

* optimize readme

* update

* add eval.yaml

README.md

@@ -280,7 +280,7 @@ python setup.py install
 答：这是因为量化后保存的参数是虽然是int8范围，但是类型是float。这是因为Paddle训练前向默认的Kernel不支持INT8 Kernel实现，只有Paddle Inference TensorRT的推理才支持量化推理加速。为了方便量化后验证量化精度，使用Paddle训练前向能加载此模型，默认保存的Float32类型权重，体积没有发生变换。
 
 
-#### 2. macOS + Python3.9环境下, 安装出错, "command 'swig' failed"
+#### 2. macOS + Python3.9环境或者Windows环境下, 安装出错, "command 'swig' failed"
 
 答: 请参考https://github.com/PaddlePaddle/PaddleSlim/issues/1258
 

example/auto_compression/README.md

@@ -35,7 +35,7 @@ wget https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/inference/MobileN
 tar -xf MobileNetV1_infer.tar
 # 下载ImageNet小型数据集
 wget https://sys-p0.bj.bcebos.com/slim_ci/ILSVRC2012_data_demo.tar.gz
-tar xf ILSVRC2012_data_demo.tar.gz
+tar -xf ILSVRC2012_data_demo.tar.gz
 ```
 
 - 2.运行
@@ -77,17 +77,66 @@ ac = AutoCompression(
     model_dir="./MobileNetV1_infer",
     model_filename="inference.pdmodel",
     params_filename="inference.pdiparams",
-    save_dir="output",
-    config={'Quantization': {}, "HyperParameterOptimization": {'max_quant_count': 5}},
+    save_dir="MobileNetV1_quant",
+    config={'Quantization': {}, "HyperParameterOptimization": {'ptq_algo': ['avg'], 'max_quant_count': 3}},
     train_dataloader=train_loader,
-    eval_dataloader=train_loader)  # eval_function to verify accuracy
+    eval_dataloader=train_loader)
 ac.compress()
 ```
 
+- 3.测试精度
+
+测试压缩前模型的精度:
+```shell
+CUDA_VISIBLE_DEVICES=0 python ./image_classification/eval.py
+### Eval Top1: 0.7171724759615384
+```
+
+测试量化模型的精度:
+```shell
+CUDA_VISIBLE_DEVICES=0 python ./image_classification/eval.py --model_dir='MobileNetV1_quant'
+### Eval Top1: 0.7166466346153846
+```
+
+量化后模型的精度相比量化前的模型几乎精度无损，由于是使用的超参搜索的方法来选择的量化参数，所以每次运行得到的量化模型精度会有些许波动。
+
+- 4.推理速度测试
+量化模型速度的测试依赖推理库的支持，所以确保安装的是带有TensorRT的PaddlePaddle。以下示例和展示的测试结果是基于Tesla V100、CUDA 10.2、python3.7得到的。
+
+使用以下指令查看本地cuda版本，并且在[下载链接](https://paddleinference.paddlepaddle.org.cn/master/user_guides/download_lib.html#python)中下载对应cuda版本和对应python版本的paddlepaddle安装包。
+```shell
+cat /usr/local/cuda/version.txt ### CUDA Version 10.2.89
+### 10.2.89 为cuda版本号，可以根据这个版本号选择需要安装的带有TensorRT的PaddlePaddle安装包。
+```
+
+安装下载的whl包：
+```
+### 这里通过wget下载到的是python3.7、cuda10.2的PaddlePaddle安装包，若您的环境和示例环境不同，请依赖您自己机器的环境下载对应的安装包，否则运行示例代码会报错。
+wget https://paddle-inference-lib.bj.bcebos.com/2.3.0/python/Linux/GPU/x86-64_gcc8.2_avx_mkl_cuda10.2_cudnn8.1.1_trt7.2.3.4/paddlepaddle_gpu-2.3.0-cp37-cp37m-linux_x86_64.whl
+pip install paddlepaddle_gpu-2.3.0-cp37-cp37m-linux_x86_64.whl --force-reinstall
+```
+
+测试FP32模型的速度
+```
+python ./image_classification/infer.py
+### using tensorrt FP32	batch size: 1 time(ms): 0.6140608787536621
+```
+
+测试FP16模型的速度
+```
+python ./image_classification/infer.py --use_fp16=True
+### using tensorrt FP16	batch size: 1 time(ms): 0.5795984268188477
+```
+
+测试INT8模型的速度
+```
+python ./image_classification/infer.py --model_dir=./MobileNetV1_quant/ --use_int8=True
+### using tensorrt INT8 batch size: 1 time(ms): 0.5213963985443115
+```
+
 **提示：**
-- DataLoader传入的数据集是待压缩模型所用的数据集，DataLoader继承自`paddle.io.DataLoader`。
-- 如无需验证自动化压缩过程中模型的精度，`eval_callback`可不传入function，程序会自动根据损失来选择最优模型。
-- 自动化压缩Config中定义量化、蒸馏、剪枝等压缩算法会合并执行，压缩策略有：量化+蒸馏，剪枝+蒸馏等等。
+- DataLoader传入的数据集是待压缩模型所用的数据集，DataLoader继承自`paddle.io.DataLoader`。可以直接使用模型套件中的DataLoader，或者根据[paddle.io.DataLoader](https://www.paddlepaddle.org.cn/documentation/docs/zh/api/paddle/io/DataLoader_cn.html#dataloader)自定义所需要的DataLoader。
+- 自动化压缩Config中定义量化、蒸馏、剪枝等压缩算法会合并执行，压缩策略有：量化+蒸馏，剪枝+蒸馏等等。示例中选择的配置为离线量化超参搜索。
 - 如果要压缩的模型参数是存储在各自分离的文件中，需要先通过[convert.py](./convert.py) 脚本将其保存成一个单独的二进制文件。
 
 ## 应用示例

example/auto_compression/image_classification/README.md

@@ -122,9 +122,9 @@ python infer.py --config_path="configs/infer.yaml"
 ```
 
 在配置文件```configs/infer.yaml```中有以下字段用于配置预测参数：
-- ```inference_model_dir```：inference 模型文件所在目录，该目录下需要有文件 .pdmodel 和 .pdiparams 两个文件
-- ```model_filename```：inference_model_dir文件夹下的模型文件名称
-- ```params_filename```：inference_model_dir文件夹下的参数文件名称
+- ```model_dir```：inference 模型文件所在目录，该目录下需要有文件 .pdmodel 和 .pdiparams 两个文件
+- ```model_filename```：model_dir文件夹下的模型文件名称
+- ```params_filename```：model_dir文件夹下的参数文件名称
 - ```batch_size```：预测一个batch的大小
 - ```image_size```：输入图像的大小
 - ```use_tensorrt```：是否使用 TesorRT 预测引擎

example/auto_compression/image_classification/configs/eval.yaml

+model_dir: './MobileNetV1_infer'
+model_filename: 'inference.pdmodel'
+params_filename: "inference.pdiparams"
+batch_size: 128
+data_dir: './ILSVRC2012_data_demo/ILSVRC2012/'
+img_size: 224
+resize_size: 256

example/auto_compression/image_classification/configs/infer.yaml

-inference_model_dir: "./MobileNetV1_infer"
+model_dir: "./MobileNetV1_infer"
 model_filename: "inference.pdmodel"
 params_filename: "inference.pdiparams"
 batch_size: 1

example/auto_compression/image_classification/eval.py

@@ -31,9 +31,13 @@ def argsparser():
     parser.add_argument(
         '--config_path',
         type=str,
-        default=None,
-        help="path of compression strategy config.",
-        required=True)
+        default='./image_classification/configs/eval.yaml',
+        help="path of compression strategy config.")
+    parser.add_argument(
+        '--model_dir',
+        type=str,
+        default='./MobileNetV1_infer',
+        help='model directory')
     return parser
 
 
@@ -92,19 +96,20 @@ def eval():
     return result[0]
 
 
-def main():
+def main(args):
     global global_config
-    all_config = load_slim_config(args.config_path)
-    assert "Global" in all_config, f"Key 'Global' not found in config file. \n{all_config}"
-    global_config = all_config["Global"]
+    global_config = load_slim_config(args.config_path)
 
     global data_dir
     data_dir = global_config['data_dir']
+    if args.model_dir != global_config['model_dir']:
+        global_config['model_dir'] = args.model_dir
 
     global img_size, resize_size
-    img_size = global_config['img_size'] if 'img_size' in global_config else 224
-    resize_size = global_config[
-        'resize_size'] if 'resize_size' in global_config else 256
+    img_size = int(global_config[
+        'img_size']) if 'img_size' in global_config else 224
+    resize_size = int(global_config[
+        'resize_size']) if 'resize_size' in global_config else 256
 
     result = eval()
     print('Eval Top1:', result)
@@ -114,4 +119,4 @@ if __name__ == '__main__':
     paddle.enable_static()
     parser = argsparser()
     args = parser.parse_args()
-    main()
+    main(args)

example/auto_compression/image_classification/infer.py

@@ -30,8 +30,17 @@ def argsparser():
     parser.add_argument(
         '--config_path',
         type=str,
-        default='configs/infer.yaml',
+        default='./image_classification/configs/infer.yaml',
         help='config file path')
+    parser.add_argument(
+        '--model_dir',
+        type=str,
+        default='./MobileNetV1_infer',
+        help='model directory')
+    parser.add_argument(
+        '--use_fp16', type=bool, default=False, help='Whether to use fp16')
+    parser.add_argument(
+        '--use_int8', type=bool, default=False, help='Whether to use int8')
     return parser
 
 
@@ -53,7 +62,7 @@ class Predictor(object):
             output_names[0])
 
     def create_paddle_predictor(self):
-        inference_model_dir = self.config['inference_model_dir']
+        inference_model_dir = self.config['model_dir']
         model_file = os.path.join(inference_model_dir,
                                   self.config['model_filename'])
         params_file = os.path.join(inference_model_dir,
@@ -110,6 +119,7 @@ class Predictor(object):
             time.sleep(0.01)  # sleep for T4 GPU
 
         fp_message = "FP16" if config['use_fp16'] else "FP32"
+        fp_message = "INT8" if config['use_int8'] else fp_message
         trt_msg = "using tensorrt" if config[
             'use_tensorrt'] else "not using tensorrt"
         print("{0}\t{1}\tbatch size: {2}\ttime(ms): {3}".format(
@@ -121,5 +131,11 @@ if __name__ == "__main__":
     parser = argsparser()
     args = parser.parse_args()
     config = load_config(args.config_path)
+    if args.model_dir != config['model_dir']:
+        config['model_dir'] = args.model_dir
+    if args.use_fp16 != config['use_fp16']:
+        config['use_fp16'] = args.use_fp16
+    if args.use_int8 != config['use_int8']:
+        config['use_int8'] = args.use_int8
     predictor = Predictor(config)
     predictor.predict()

example/auto_compression/semantic_segmentation/README.md

@@ -42,16 +42,6 @@
   - 软件：CUDA 11.0, cuDNN 8.0, TensorRT 8.0
   - 测试配置：batch_size: 40, max_seq_len: 128
 
-- PP-HumanSeg-Lite数据集
-
-  - 数据集：AISegment + PP-HumanSeg14K + 内部自建数据集。其中 AISegment 是开源数据集，可从[链接](https://github.com/aisegmentcn/matting_human_datasets)处获取；PP-HumanSeg14K 是 PaddleSeg 自建数据集，可从[官方渠道](https://github.com/PaddlePaddle/PaddleSeg/blob/release/2.5/contrib/PP-HumanSeg/paper.md#pp-humanseg14k-a-large-scale-teleconferencing-video-dataset)获取；内部数据集不对外公开。
-  - 示例数据集: 用于快速跑通人像分割的压缩和推理流程, 不能用该数据集复现 benckmark 表中的压缩效果。 [下载链接](https://paddleseg.bj.bcebos.com/humanseg/data/mini_supervisely.zip)
-
-- PP-Liteseg，HRNet，UNet，Deeplabv3-ResNet50数据集
-
-  - cityscapes: 请从[cityscapes官网](https://www.cityscapes-dataset.com/login/)下载完整数据
-  - 示例数据集: cityscapes数据集的一个子集，用于快速跑通压缩和推理流程，不能用该数据集复现 benchmark 表中的压缩效果。[下载链接](https://bj.bcebos.com/v1/paddle-slim-models/data/mini_cityscapes/mini_cityscapes.tar)
-
 下面将以开源数据集为例介绍如何对PP-HumanSeg-Lite进行自动压缩。
 
 ## 3. 自动压缩流程
@@ -87,14 +77,25 @@ pip install paddleseg
 
 开发者可下载开源数据集 (如[AISegment](https://github.com/aisegmentcn/matting_human_datasets)) 或自定义语义分割数据集。请参考[PaddleSeg数据准备文档](https://github.com/PaddlePaddle/PaddleSeg/blob/release/2.5/docs/data/marker/marker_cn.md)来检查、对齐数据格式即可。
 
-可以通过以下命令下载人像分割示例数据:
+本示例使用示例开源数据集 AISegment 数据集为例介绍如何对PP-HumanSeg-Lite进行自动压缩。示例中的数据集仅用于快速跑通自动压缩流程，并不能复现出 benckmark 表中的压缩效果。
 
+可以通过以下命令下载人像分割示例数据:
 ```shell
-cd ./data
-python download_data.py mini_humanseg
-
+python ./data/download_data.py mini_humanseg
+### 下载后的数据位置为 ./data/humanseg/
 ```
 
+** 提示: **
+- PP-HumanSeg-Lite压缩过程使用的数据集
+
+  - 数据集：AISegment + PP-HumanSeg14K + 内部自建数据集。其中 AISegment 是开源数据集，可从[链接](https://github.com/aisegmentcn/matting_human_datasets)处获取；PP-HumanSeg14K 是 PaddleSeg 自建数据集，可从[官方渠道](https://github.com/PaddlePaddle/PaddleSeg/blob/release/2.5/contrib/PP-HumanSeg/paper.md#pp-humanseg14k-a-large-scale-teleconferencing-video-dataset)获取；内部数据集不对外公开。
+  - 示例数据集: 用于快速跑通人像分割的压缩和推理流程, 不能用该数据集复现 benckmark 表中的压缩效果。 [下载链接](https://paddleseg.bj.bcebos.com/humanseg/data/mini_supervisely.zip)
+
+- PP-Liteseg，HRNet，UNet，Deeplabv3-ResNet50数据集
+
+  - cityscapes: 请从[cityscapes官网](https://www.cityscapes-dataset.com/login/)下载完整数据
+  - 示例数据集: cityscapes数据集的一个子集，用于快速跑通压缩和推理流程，不能用该数据集复现 benchmark 表中的压缩效果。[下载链接](https://bj.bcebos.com/v1/paddle-slim-models/data/mini_cityscapes/mini_cityscapes.tar)
+
 #### 3.3 准备预测模型
 
 预测模型的格式为：`model.pdmodel` 和 `model.pdiparams`两个，带`pdmodel`的是模型文件，带`pdiparams`后缀的是权重文件。
@@ -112,77 +113,40 @@ tar -xzf ppseg_lite_portrait_398x224_with_softmax.tar.gz
 
 #### 3.4 自动压缩并产出模型
 
-自动压缩示例通过run.py脚本启动，会使用接口```paddleslim.auto_compression.AutoCompression```对模型进行自动压缩。首先要配置config文件中模型路径、数据集路径、蒸馏、量化、稀疏化和训练等部分的参数，配置完成后便可对模型进行非结构化稀疏、蒸馏和量化、蒸馏。
+自动压缩示例通过run.py脚本启动，会使用接口 ```paddleslim.auto_compression.AutoCompression``` 对模型进行自动压缩。首先要配置config文件中模型路径、数据集路径、蒸馏、量化、稀疏化和训练等部分的参数，配置完成后便可对模型进行非结构化稀疏、蒸馏和量化、蒸馏。
 
-当只设置训练参数，并传入``deploy_hardware``字段时，将自动搜索压缩策略进行压缩。以骁龙710（SD710）为部署硬件，进行自动压缩的运行命令如下：
+当只设置训练参数，并在config文件中 ```Global``` 配置中传入 ```deploy_hardware``` 字段时，将自动搜索压缩策略进行压缩。以骁龙710（SD710）为部署硬件，进行自动压缩的运行命令如下：
 
 ```shell
 # 单卡启动
 export CUDA_VISIBLE_DEVICES=0
-python run.py \
-    --model_dir='./ppseg_lite_portrait_398x224_with_softmax' \
-    --model_filename='model.pdmodel' \
-    --params_filename='model.pdiparams' \
-    --save_dir='./save_model' \
-    --strategy_config='configs/pp_humanseg/pp_humanseg_auto.yaml' \
-    --dataset_config='configs/dataset/humanseg_dataset.yaml' \
-    --deploy_hardware='SD710'
+python run.py --config_path='./configs/pp_humanseg/pp_humanseg_auto.yaml' --save_dir='./save_compressed_model'
 
 # 多卡启动
 export CUDA_VISIBLE_DEVICES=0,1
-python -m paddle.distributed.launch run.py \
-    --model_dir='./ppseg_lite_portrait_398x224_with_softmax' \
-    --model_filename='model.pdmodel' \
-    --params_filename='model.pdiparams' \
-    --save_dir='./save_model' \
-    --strategy_config='configs/pp_humanseg/pp_humanseg_auto.yaml' \
-    --dataset_config='configs/dataset/humanseg_dataset.yaml' \
-    --deploy_hardware='SD710'
+python -m paddle.distributed.launch run.py --config_path='./configs/pp_humanseg/pp_humanseg_auto.yaml' --save_dir='./save_compressed_model'
 ```
+
 - 自行配置稀疏参数进行非结构化稀疏和蒸馏训练，配置参数含义详见[自动压缩超参文档](https://github.com/PaddlePaddle/PaddleSlim/blob/27dafe1c722476f1b16879f7045e9215b6f37559/demo/auto_compression/hyperparameter_tutorial.md)。具体命令如下所示：
 ```shell
 # 单卡启动
 export CUDA_VISIBLE_DEVICES=0
-python run.py \
-    --model_dir='./ppseg_lite_portrait_398x224_with_softmax' \
-    --model_filename='model.pdmodel' \
-    --params_filename='model.pdiparams' \
-    --save_dir='./save_model' \
-    --strategy_config='configs/pp_humanseg/pp_humanseg_sparse.yaml' \
-    --dataset_config='configs/dataset/humanseg_dataset.yaml'
+python run.py --config_path='./configs/pp_humanseg/pp_humanseg_sparse.yaml' --save_dir='./save_sparse_model'
 
 # 多卡启动
 export CUDA_VISIBLE_DEVICES=0,1
-python -m paddle.distributed.launch run.py \
-    --model_dir='./ppseg_lite_portrait_398x224_with_softmax' \
-    --model_filename='model.pdmodel' \
-    --params_filename='model.pdiparams' \
-    --save_dir='./save_model' \
-    --strategy_config='configs/pp_humanseg/pp_humanseg_sparse.yaml' \
-    --dataset_config='configs/dataset/humanseg_dataset.yaml'
+python -m paddle.distributed.launch run.py --config_path='./configs/pp_humanseg/pp_humanseg_sparse.yaml' --save_dir='./save_sparse_model'
 ```
 
 - 自行配置量化参数进行量化和蒸馏训练，配置参数含义详见[自动压缩超参文档](https://github.com/PaddlePaddle/PaddleSlim/blob/27dafe1c722476f1b16879f7045e9215b6f37559/demo/auto_compression/hyperparameter_tutorial.md)。具体命令如下所示：
 ```shell
 # 单卡启动
 export CUDA_VISIBLE_DEVICES=0
-python run.py \
-    --model_dir='./ppseg_lite_portrait_398x224_with_softmax' \
-    --model_filename='model.pdmodel' \
-    --params_filename='model.pdiparams' \
-    --save_dir='./save_model' \
-    --strategy_config='configs/pp_humanseg/pp_humanseg_qat.yaml' \
-    --dataset_config='configs/dataset/humanseg_dataset.yaml'
+python run.py --config_path='./configs/pp_humanseg/pp_humanseg_qat.yaml' --save_dir='./save_quant_model'
 
 # 多卡启动
 export CUDA_VISIBLE_DEVICES=0,1
-python -m paddle.distributed.launch run.py \
-    --model_dir='./ppseg_lite_portrait_398x224_with_softmax' \
-    --model_filename='model.pdmodel' \
-    --params_filename='model.pdiparams' \
-    --save_dir='./save_model' \
-    --strategy_config='configs/pp_humanseg/pp_humanseg_qat.yaml' \
-    --dataset_config='configs/dataset/humanseg_dataset.yaml'
+python -m paddle.distributed.launch run.py --config_path='./configs/pp_humanseg/pp_humanseg_qat.yaml' --save_dir='./save_quant_model'
 ```
 
 压缩完成后会在`save_dir`中产出压缩好的预测模型，可直接预测部署。

example/auto_compression/semantic_segmentation/configs/deeplabv3/deeplabv3_qat.yaml

 Global:
   reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml
+  model_dir: ./RES-paddle2-Deeplabv3-ResNet50
+  model_filename: model
+  params_filename: params
 
 Distillation:
   alpha: 1.0

example/auto_compression/semantic_segmentation/configs/hrnet/hrnet_qat.yaml

 Global:
   reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml
+  model_dir: ./RES-paddle2-HRNetW18-Seg
+  model_filename: model
+  params_filename: params
 
 Distillation:
   alpha: 1.0

example/auto_compression/semantic_segmentation/configs/hrnet/hrnet_sparse.yaml

 Global:
   reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml
+  model_dir: ./RES-paddle2-HRNetW18-Seg
+  model_filename: model
+  params_filename: params
 
 Distillation:
   alpha: 1.0

example/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_auto.yaml

+Global:
+  reader_config: configs/dataset/humanseg_dataset.yaml
+  model_dir: ./ppseg_lite_portrait_398x224_with_softmax
+  model_filename: model.pdmodel
+  params_filename: model.pdiparams
+  deploy_hardware: SD710
+
 TrainConfig:
   epochs: 14
   eval_iter: 400

example/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_qat.yaml

+Global:
+  reader_config: configs/dataset/humanseg_dataset.yaml
+  model_dir: ./ppseg_lite_portrait_398x224_with_softmax
+  model_filename: model.pdmodel
+  params_filename: model.pdiparams
+
 Distillation:
   alpha: 1.0
   loss: l2

example/auto_compression/semantic_segmentation/configs/pp_humanseg/pp_humanseg_sparse.yaml

+Global:
+  reader_config: configs/dataset/humanseg_dataset.yaml
+  model_dir: ./ppseg_lite_portrait_398x224_with_softmax
+  model_filename: model.pdmodel
+  params_filename: model.pdiparams
+
 Distillation:
   alpha: 1.0
   loss: l2

example/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_auto.yaml

+Global:
+  reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml
+  model_dir: ./RES-paddle2-PPLIteSegSTDC1
+  model_filename: model
+  params_filename: params
+
 TrainConfig:
  epochs: 14
  eval_iter: 90

example/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_qat.yaml

+Global:
+  reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml
+  model_dir: ./RES-paddle2-PPLIteSegSTDC1
+  model_filename: model
+  params_filename: params
+
 Distillation:
   alpha: 1.0
   loss: l2

example/auto_compression/semantic_segmentation/configs/pp_liteseg/pp_liteseg_sparse.yaml

+Global:
+  reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml
+  model_dir: ./RES-paddle2-PPLIteSegSTDC1
+  model_filename: model
+  params_filename: params
+
 Distillation:
   alpha: 1.0
   loss: l2

example/auto_compression/semantic_segmentation/configs/unet/unet_channel_prune.yaml

 Global:
   reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml
+  model_dir: ./RES-paddle2-UNet
+  model_filename: model
+  params_filename: params
 
 Distillation:
   alpha: 1.0

example/auto_compression/semantic_segmentation/configs/unet/unet_qat.yaml

 Global:
   reader_config: configs/dataset/cityscapes_1024x512_scale1.0.yml
+  model_dir: ./RES-paddle2-UNet
+  model_filename: model
+  params_filename: params
 
 Distillation:
   alpha: 1.0

example/auto_compression/semantic_segmentation/run.py

@@ -21,48 +21,24 @@ from paddleseg.cvlibs import Config as PaddleSegDataConfig
 from paddleseg.utils import worker_init_fn
 
 from paddleslim.auto_compression import AutoCompression
+from paddleslim.auto_compression.config_helpers import load_config as load_slim_config
 from paddleseg.core.infer import reverse_transform
 from paddleseg.utils import metrics
 
 
-def parse_args():
-    parser = argparse.ArgumentParser(description='Model training')
+def argsparser():
+    parser = argparse.ArgumentParser(description=__doc__)
     parser.add_argument(
-        '--model_dir',
+        '--config_path',
         type=str,
         default=None,
-        help="inference model directory.")
-    parser.add_argument(
-        '--model_filename',
-        type=str,
-        default=None,
-        help="inference model filename.")
-    parser.add_argument(
-        '--params_filename',
-        type=str,
-        default=None,
-        help="inference params filename.")
+        help="path of compression strategy config.")
     parser.add_argument(
         '--save_dir',
         type=str,
         default=None,
         help="directory to save compressed model.")
-    parser.add_argument(
-        '--strategy_config',
-        type=str,
-        default=None,
-        help="path of compression strategy config.")
-    parser.add_argument(
-        '--dataset_config',
-        type=str,
-        default=None,
-        help="path of dataset config.")
-    parser.add_argument(
-        '--deploy_hardware',
-        type=str,
-        default=None,
-        help="The hardware you want to deploy.")
-    return parser.parse_args()
+    return parser
 
 
 def eval_function(exe, compiled_test_program, test_feed_names, test_fetch_list):
@@ -141,13 +117,15 @@ def reader_wrapper(reader):
     return gen
 
 
-if __name__ == '__main__':
+def main(args):
+    all_config = load_slim_config(args.config_path)
+    assert "Global" in all_config, f"Key 'Global' not found in config file. \n{all_config}"
+    config = all_config["Global"]
+
     rank_id = paddle.distributed.get_rank()
     place = paddle.CUDAPlace(rank_id)
-    args = parse_args()
-    paddle.enable_static()
     # step1: load dataset config and create dataloader
-    data_cfg = PaddleSegDataConfig(args.dataset_config)
+    data_cfg = PaddleSegDataConfig(config['reader_config'])
     train_dataset = data_cfg.train_dataset
     eval_dataset = data_cfg.val_dataset
     batch_sampler = paddle.io.DistributedBatchSampler(
@@ -166,19 +144,24 @@ if __name__ == '__main__':
 
     nranks = paddle.distributed.get_world_size()
     rank_id = paddle.distributed.get_rank()
-    if nranks > 1 and rank_id != 0:
-        eval_function = None
 
     # step2: create and instance of AutoCompression
     ac = AutoCompression(
-        model_dir=args.model_dir,
-        model_filename=args.model_filename,
-        params_filename=args.params_filename,
+        model_dir=config['model_dir'],
+        model_filename=config['model_filename'],
+        params_filename=config['params_filename'],
         save_dir=args.save_dir,
-        config=args.strategy_config,
+        config=all_config,
         train_dataloader=train_dataloader,
-        eval_callback=eval_function,
-        deploy_hardware=args.deploy_hardware)
+        eval_callback=eval_function if nranks > 1 and rank_id != 0 else None,
+        deploy_hardware=config.get('deploy_hardware') or None)
 
     # step3: start the compression job
     ac.compress()
+
+
+if __name__ == '__main__':
+    paddle.enable_static()
+    parser = argsparser()
+    args = parser.parse_args()
+    main(args)

paddleslim/auto_compression/compressor.py

@@ -151,8 +151,9 @@ class AutoCompression:
         self.train_dataloader = wrap_dataloader(train_dataloader,
                                                 self.feed_vars)
         self.eval_dataloader = wrap_dataloader(eval_dataloader, self.feed_vars)
-        if eval_dataloader is None:
-            eval_dataloader = self._get_eval_dataloader(self.train_dataloader)
+        if self.eval_dataloader is None:
+            self.eval_dataloader = self._get_eval_dataloader(
+                self.train_dataloader)
 
         self.target_speedup = target_speedup
         self.eval_function = eval_callback
@@ -465,10 +466,10 @@ class AutoCompression:
                     'train_config must has `epochs` or `train_iter` field.')
             config_dict['gmp_config'] = {
                 'stable_iterations': 0,
-                'pruning_iterations': 0.45 * total_iters,
-                'tunning_iterations': 0.45 * total_iters,
+                'pruning_iterations': max(0.45 * total_iters, 30),
+                'tunning_iterations': max(0.45 * total_iters, 30),
                 'resume_iteration': -1,
-                'pruning_steps': 100,
+                'pruning_steps': 100 if (0.45 * total_iters) > 1000 else 1,
                 'initial_ratio': 0.15,
             }
         ### add prune program