fix scripts and docs (#3465)

PaddleSlim/classification/distillation/README.md

@@ -5,6 +5,8 @@
 ## 概述
 
 该示例使用PaddleSlim提供的[蒸馏策略]([https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/tutorial.md#3-%E8%92%B8%E9%A6%8F](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/tutorial.md#3-蒸馏))对分类模型进行知识蒸馏。
+>本文默认使用ILSVRC2012数据集，数据集存放在`models/PaddleSlim/data/`路径下, 可以参考[数据准备](https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/image_classification#数据准备)在执行训练脚本run.sh前配置好您的数据集
+
 在阅读该示例前，建议您先了解以下内容：
 
 - [分类模型的常规训练方法](https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/image_classification)

PaddleSlim/classification/distillation/run.sh

@@ -2,9 +2,6 @@
 
 # download pretrain model
 root_url="http://paddle-imagenet-models-name.bj.bcebos.com"
-MobileNetV1="MobileNetV1_pretrained.tar"
-MobileNetV2="MobileNetV2_pretrained.tar"
-ResNet34="ResNet34_pretrained.tar"
 ResNet50="ResNet50_pretrained.tar"
 pretrain_dir='../pretrain'
 
@@ -14,15 +11,6 @@ fi
 
 cd ${pretrain_dir}
 
-if [ ! -f ${MobileNetV2} ]; then
-    wget ${root_url}/${MobileNetV2}
-    tar xf ${MobileNetV2}
-fi
-
-if [ ! -f ${ResNet34} ]; then
-    wget ${root_url}/${ResNet34}
-    tar xf ${ResNet34}
-fi
 if [ ! -f ${ResNet50} ]; then
     wget ${root_url}/${ResNet50}
     tar xf ${ResNet50}
@@ -62,7 +50,7 @@ for files in $(ls res50_*)
 done
 cd -
 
-# for mobilenet_v2 distillation
+## for mobilenet_v2 distillation
 #cd ${pretrain_dir}/ResNet50_pretrained
 #for files in $(ls res50_*)
 #    do mv $files ${files#*_}
@@ -86,7 +74,7 @@ cd -
 #done
 #cd -
 
-# for resnet34 distillation
+## for resnet34 distillation
 #cd ${pretrain_dir}/ResNet50_pretrained
 #for files in $(ls res50_*)
 #    do mv $files ${files#*_}
@@ -96,16 +84,7 @@ cd -
 #done
 #cd -
 #
-#cd ${pretrain_dir}/ResNet34_pretrained
-#for files in $(ls res34_*)
-#    do mv $files ${files#*_}
-#done
-#for files in $(ls *)
-#    do mv $files "res34_"$files
-#done
-#cd -
-#
-#python compress.py \
+#python -u compress.py \
 #--model "ResNet34" \
 #--teacher_model "ResNet50" \
 #--teacher_pretrained_model ../pretrain/ResNet50_pretrained \
@@ -118,10 +97,3 @@ cd -
 #    do mv $files ${files#*_}
 #done
 #cd -
-#
-#cd ${pretrain_dir}/ResNet34_pretrained
-#for files in $(ls res34_*)
-#    do mv $files ${files#*_}
-#done
-#cd -
-

PaddleSlim/classification/pruning/README.md

@@ -5,6 +5,8 @@
 ## 概述
 
 该示例使用PaddleSlim提供的[卷积通道剪裁压缩策略](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/tutorial.md#2-%E5%8D%B7%E7%A7%AF%E6%A0%B8%E5%89%AA%E8%A3%81%E5%8E%9F%E7%90%86)对分类模型进行压缩。
+>本文默认使用ILSVRC2012数据集，数据集存放在`models/PaddleSlim/data/`路径下, 可以参考[数据准备](https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/image_classification#数据准备)在执行训练脚本run.sh前配置好您的数据集
+
 在阅读该示例前，建议您先了解以下内容：
 
 - [分类模型的常规训练方法](https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/image_classification)

PaddleSlim/classification/pruning/run.sh

+#!/usr/bin/env bash
+
 export CUDA_VISIBLE_DEVICES=0
 
-nohup python compress.py \
+# download pretrain model
+root_url="http://paddle-imagenet-models-name.bj.bcebos.com"
+MobileNetV1="MobileNetV1_pretrained.tar"
+MobileNetV2="MobileNetV2_pretrained.tar"
+ResNet50="ResNet50_pretrained.tar"
+pretrain_dir='../pretrain'
+
+if [ ! -d ${pretrain_dir} ]; then
+  mkdir ${pretrain_dir}
+fi
+
+cd ${pretrain_dir}
+
+if [ ! -f ${MobileNetV1} ]; then
+    wget ${root_url}/${MobileNetV1}
+    tar xf ${MobileNetV1}
+fi
+
+if [ ! -f ${MobileNetV2} ]; then
+    wget ${root_url}/${MobileNetV2}
+    tar xf ${MobileNetV2}
+fi
+
+if [ ! -f ${ResNet50} ]; then
+    wget ${root_url}/${ResNet50}
+    tar xf ${ResNet50}
+fi
+
+cd -
+
+nohup python -u compress.py \
 --model "MobileNet" \
---use_gpu 0 \
---batch_size 1 \
+--use_gpu 1 \
+--batch_size 256 \
 --pretrained_model ../pretrain/MobileNetV1_pretrained \
 --config_file "./configs/mobilenet_v1.yaml" \
 > mobilenet_v1.log 2>&1 &
 tailf mobilenet_v1.log
 
 # for compression of mobilenet_v2
-#nohup python compress.py \
+#nohup python -u compress.py \
 #--model "MobileNetV2" \
-#--use_gpu 0 \
-#--batch_size 1 \
+#--use_gpu 1 \
+#--batch_size 256 \
 #--pretrained_model ../pretrain/MobileNetV2_pretrained \
 #--config_file "./configs/mobilenet_v2.yaml" \
 #> mobilenet_v2.log 2>&1 &
 #tailf mobilenet_v2.log
 
 
-# for compression of resnet50
-#python compress.py \
+## for compression of resnet50
+#python -u compress.py \
 #--model "ResNet50" \
-#--use_gpu 0 \
-#--batch_size 1 \
+#--use_gpu 1 \
+#--batch_size 256 \
 #--pretrained_model ../pretrain/ResNet50_pretrained \
 #--config_file "./configs/resnet50.yaml" \
 #> resnet50.log 2>&1 &
 #tailf resnet50.log
-

PaddleSlim/classification/quantization/README.md

@@ -5,6 +5,8 @@
 ## 概述
 
 该示例使用PaddleSlim提供的[量化压缩策略](https://github.com/PaddlePaddle/models/blob/develop/PaddleSlim/docs/tutorial.md#1-quantization-aware-training%E9%87%8F%E5%8C%96%E4%BB%8B%E7%BB%8D)对分类模型进行压缩。
+>本文默认使用ILSVRC2012数据集，数据集存放在`models/PaddleSlim/data/`路径下, 可以参考[数据准备](https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/image_classification#数据准备)在执行训练脚本run.sh前配置好您的数据集
+
 在阅读该示例前，建议您先了解以下内容：
 
 - [分类模型的常规训练方法](https://github.com/PaddlePaddle/models/tree/develop/PaddleCV/image_classification)

PaddleSlim/classification/quantization/run.sh

 #!/usr/bin/env bash
 
-#export CUDA_VISIBLE_DEVICES=0
+# download pretrain model
+root_url="http://paddle-imagenet-models-name.bj.bcebos.com"
+MobileNetV1="MobileNetV1_pretrained.tar"
+MobileNetV2="MobileNetV2_pretrained.tar"
+ResNet50="ResNet50_pretrained.tar"
+pretrain_dir='../pretrain'
 
-# for quantization for mobilenet_v1
-#python compress.py \
-#    --model "MobileNet" \
-#    --use_gpu 1 \
-#    --batch_size 32 \
-#    --pretrained_model ../pretrain/MobileNetV1_pretrained \
-#    --config_file "./configs/mobilenet_v1.yaml" \
-#> mobilenet_v1.log 2>&1 &
-#tailf mobilenet_v1.log
+if [ ! -d ${pretrain_dir} ]; then
+  mkdir ${pretrain_dir}
+fi
+
+cd ${pretrain_dir}
+
+if [ ! -f ${MobileNetV1} ]; then
+    wget ${root_url}/${MobileNetV1}
+    tar xf ${MobileNetV1}
+fi
+
+if [ ! -f ${MobileNetV2} ]; then
+    wget ${root_url}/${MobileNetV2}
+    tar xf ${MobileNetV2}
+fi
+
+if [ ! -f ${ResNet50} ]; then
+    wget ${root_url}/${ResNet50}
+    tar xf ${ResNet50}
+fi
+
+cd -
+
+# enable GC strategy
+export FLAGS_fast_eager_deletion_mode=1
+export FLAGS_eager_delete_tensor_gb=0.0
+
+export CUDA_VISIBLE_DEVICES=0
 
-# for quantization of mobilenet_v2
-# python compress.py \
+## for quantization for mobilenet_v1
+python -u compress.py \
+    --model "MobileNet" \
+    --use_gpu 1 \
+    --batch_size 32 \
+    --pretrained_model ../pretrain/MobileNetV1_pretrained \
+    --config_file "./configs/mobilenet_v1.yaml" \
+> mobilenet_v1.log 2>&1 &
+tailf mobilenet_v1.log
+
+## for quantization of mobilenet_v2
+#python -u compress.py \
 #    --model "MobileNetV2" \
 #    --use_gpu 1 \
 #    --batch_size 32 \
@@ -22,14 +56,12 @@
 #    > mobilenet_v2.log 2>&1 &
 #tailf mobilenet_v2.log
 
-
 # for compression of resnet50
-python compress.py \
-    --model "ResNet50" \
-    --use_gpu 1 \
-    --batch_size 32 \
-    --pretrained_model ../pretrain/ResNet50_pretrained \
-    --config_file "./configs/resnet50.yaml" \
-    > resnet50.log 2>&1 &
-tailf resnet50.log
-
+#python -u compress.py \
+#    --model "ResNet50" \
+#    --use_gpu 1 \
+#    --batch_size 32 \
+#    --pretrained_model ../pretrain/ResNet50_pretrained \
+#    --config_file "./configs/resnet50.yaml" \
+#    > resnet50.log 2>&1 &
+#tailf resnet50.log