Fix Auto_seg's demo (#1124)

* Add dataset download link.

* Fix some bugs.

* Fix some bugs.

* Fix docs.

demo/auto_compression/hyperparameter_tutorial.md

@@ -56,7 +56,7 @@ Prune:
   # pruned_ratio: 裁剪比例
   pruned_ratio: 0.25
   # prune_params_name: 需要裁剪的参数名字
-  prune_params_name: 
+  prune_params_name:
   - conv1_weights
   # criterion: 评估一个卷积层内通道重要性所参考的指标
   criterion: l1_norm
@@ -79,12 +79,12 @@ UnstructurePrune:
     threshold: 0.001
     # gmp_config: 传入额外的训练超参用以指导GMP训练过程
     gmp_config:
-    - stable_iterations: 0
-    - pruning_iterations: 4500 # total_iters * 0.4~0.45
-    - tunning_iterations: 4500 # total_iters * 0.4~0.45
-    - resume_iteration: -1
-    - pruning_steps: 100
-    - initial_ratio: 0.15
+      stable_iterations: 0
+      pruning_iterations: 4500 # total_iters * 0.4~0.45
+      tunning_iterations: 4500 # total_iters * 0.4~0.45
+      resume_iteration: -1
+      pruning_steps: 100
+      initial_ratio: 0.15
     # prune_params_type: 用以指定哪些类型的参数参与稀疏。
     prune_params_type: conv1x1_only
     # local_sparsity: 剪裁比例（ratio）应用的范围

demo/auto_compression/semantic_segmentation/README.md

@@ -61,7 +61,7 @@ pip install paddleseg
 
 #### 3.2 准备数据集
 
-开发者可下载开源数据集或自定义语义分割数据集，例如PP-HumanSeg-Lite模型中使用的语义分割数据集[PP-HumanSeg14K](https://github.com/PaddlePaddle/PaddleSeg/blob/release/2.5/contrib/PP-HumanSeg/paper.md#pp-humanseg14k-a-large-scale-teleconferencing-video-dataset)可从官方渠道下载。
+开发者可下载开源数据集 ([AIsegment](https://github.com/aisegmentcn/matting_human_datasets)) 或自定义语义分割数据集，例如PP-HumanSeg-Lite模型中使用的语义分割数据集[PP-HumanSeg14K](https://github.com/PaddlePaddle/PaddleSeg/blob/release/2.5/contrib/PP-HumanSeg/paper.md#pp-humanseg14k-a-large-scale-teleconferencing-video-dataset)可从官方渠道下载。
 
 如果是自定义数据，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。
 
@@ -74,7 +74,7 @@ pip install paddleseg
 - 如果想快速体验，可直接下载PP-HumanSeg-Lite 的预测模型：
 
 ```shell
-wget https://paddleseg.bj.bcebos.com/dygraph/ppseg/ppseg_lite_portrait_398x224_with_softmax.tar.gz
+wget https://bj.bcebos.com/v1/paddlemodels/PaddleSlim/analysis/ppseg_lite_portrait_398x224_with_softmax.tar.gz
 tar -xzf ppseg_lite_portrait_398x224_with_softmax.tar.gz
 ```
 
@@ -87,6 +87,7 @@ tar -xzf ppseg_lite_portrait_398x224_with_softmax.tar.gz
 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：
 
 ```shell
+# 单卡启动
 export CUDA_VISIBLE_DEVICES=0
 python run.py \
     --model_dir='./ppseg_lite_portrait_398x224_with_softmax' \
@@ -95,9 +96,20 @@ python run.py \
     --save_dir='./save_model' \
     --config_path='configs/pp_humanseg_auto.yaml' \
     --deploy_hardware='SD710'
+
+# 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' \
+    --config_path='configs/pp_humanseg_auto.yaml'
+    --deploy_hardware='SD710'
 ```
 - 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' \
@@ -105,10 +117,20 @@ python run.py \
     --params_filename='model.pdiparams' \
     --save_dir='./save_model' \
     --config_path='configs/pp_humanseg_sparse_dis.yaml'
+
+# 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' \
+    --config_path='configs/pp_humanseg_sparse_dis.yaml'
 ```
 
 - 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' \
@@ -116,6 +138,15 @@ python run.py \
     --params_filename='model.pdiparams' \
     --save_dir='./save_model' \
     --config_path='configs/pp_humanseg_quant_dis.yaml'
+
+# 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' \
+    --config_path='configs/pp_humanseg_quant_dis.yaml'
 ```
 
 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。

demo/auto_compression/semantic_segmentation/configs/pp_humanseg_sparse_dis.yaml

@@ -27,7 +27,10 @@ UnstructurePrune:
 TrainConfig:
   epochs: 14
   eval_iter: 400
-  learning_rate: 5.0e-03
+  learning_rate: 
+    type: PiecewiseDecay
+    boundaries: [4500]
+    values: [0.005, 0.0005]
   optim_args:
     weight_decay: 0.0005
   optimizer: SGD