fix distributed train in detection act (#1210)

demo/auto_compression/detection/README.md

@@ -96,14 +96,19 @@ tar -xf ppyoloe_crn_l_300e_coco.tar
 #### 3.4 自动压缩并产出模型
 
 蒸馏量化自动压缩示例通过run.py脚本启动，会使用接口```paddleslim.auto_compression.AutoCompression```对模型进行自动压缩。配置config文件中模型路径、蒸馏、量化、和训练等部分的参数，配置完成后便可对模型进行量化和蒸馏。具体运行命令为：
+
+- 单卡训练：
 ```
-# 单卡
 export CUDA_VISIBLE_DEVICES=0
-# 多卡
-# export CUDA_VISIBLE_DEVICES=0,1,2,3
 python run.py --config_path=./configs/ppyoloe_l_qat_dis.yaml --save_dir='./output/'
 ```
 
+- 多卡训练：
+```
+CUDA_VISIBLE_DEVICES=0,1,2,3 python -m paddle.distributed.launch --log_dir=log --gpus 0,1,2,3 run.py \
+          --config_path=./configs/ppyoloe_l_qat_dis.yaml --save_dir='./output/'
+```
+
 #### 3.5 测试模型精度
 
 使用eval.py脚本得到模型的mAP：

demo/auto_compression/detection/run.py

@@ -124,28 +124,31 @@ def main():
                                         return_list=True)
     train_loader = reader_wrapper(train_loader, global_config['input_list'])
 
-    dataset = reader_cfg['EvalDataset']
-    global val_loader
-    val_loader = create('EvalReader')(reader_cfg['EvalDataset'],
-                                      reader_cfg['worker_num'],
-                                      return_list=True)
-    metric = None
-    if reader_cfg['metric'] == 'COCO':
-        clsid2catid = {v: k for k, v in dataset.catid2clsid.items()}
-        anno_file = dataset.get_anno()
-        metric = COCOMetric(
-            anno_file=anno_file, clsid2catid=clsid2catid, IouType='bbox')
-    elif reader_cfg['metric'] == 'VOC':
-        metric = VOCMetric(
-            label_list=dataset.get_label_list(),
-            class_num=reader_cfg['num_classes'],
-            map_type=reader_cfg['map_type'])
-    else:
-        raise ValueError("metric currently only supports COCO and VOC.")
-    global_config['metric'] = metric
-
-    if 'Evaluation' in global_config.keys() and global_config['Evaluation']:
+    if 'Evaluation' in global_config.keys() and global_config[
+            'Evaluation'] and paddle.distributed.get_rank() == 0:
         eval_func = eval_function
+        dataset = reader_cfg['EvalDataset']
+        global val_loader
+        _eval_batch_sampler = paddle.io.BatchSampler(
+            dataset, batch_size=reader_cfg['EvalReader']['batch_size'])
+        val_loader = create('EvalReader')(dataset,
+                                          reader_cfg['worker_num'],
+                                          batch_sampler=_eval_batch_sampler,
+                                          return_list=True)
+        metric = None
+        if reader_cfg['metric'] == 'COCO':
+            clsid2catid = {v: k for k, v in dataset.catid2clsid.items()}
+            anno_file = dataset.get_anno()
+            metric = COCOMetric(
+                anno_file=anno_file, clsid2catid=clsid2catid, IouType='bbox')
+        elif reader_cfg['metric'] == 'VOC':
+            metric = VOCMetric(
+                label_list=dataset.get_label_list(),
+                class_num=reader_cfg['num_classes'],
+                map_type=reader_cfg['map_type'])
+        else:
+            raise ValueError("metric currently only supports COCO and VOC.")
+        global_config['metric'] = metric
     else:
         eval_func = None
 

demo/auto_compression/pytorch_yolov5/README.md

@@ -92,14 +92,19 @@ cp -r pd_model/inference_model/ yolov5_inference_model
 #### 3.4 自动压缩并产出模型
 
 蒸馏量化自动压缩示例通过run.py脚本启动，会使用接口```paddleslim.auto_compression.AutoCompression```对模型进行自动压缩。配置config文件中模型路径、蒸馏、量化、和训练等部分的参数，配置完成后便可对模型进行量化和蒸馏。具体运行命令为：
+
+- 单卡训练：
 ```
-# 单卡
 export CUDA_VISIBLE_DEVICES=0
-# 多卡
-# export CUDA_VISIBLE_DEVICES=0,1,2,3
 python run.py --config_path=./configs/yolov5s_qat_dis.yaml --save_dir='./output/'
 ```
 
+- 多卡训练：
+```
+CUDA_VISIBLE_DEVICES=0,1,2,3 python -m paddle.distributed.launch --log_dir=log --gpus 0,1,2,3 run.py \
+          --config_path=./configs/yolov5s_qat_dis.yaml --save_dir='./output/'
+```
+
 #### 3.5 测试模型精度
 
 使用eval.py脚本得到模型的mAP：

demo/auto_compression/pytorch_yolov5/run.py

@@ -138,28 +138,31 @@ def main():
                                         return_list=True)
     train_loader = reader_wrapper(train_loader, global_config['input_list'])
 
-    dataset = reader_cfg['EvalDataset']
-    global val_loader
-    val_loader = create('EvalReader')(reader_cfg['EvalDataset'],
-                                      reader_cfg['worker_num'],
-                                      return_list=True)
-    metric = None
-    if reader_cfg['metric'] == 'COCO':
-        clsid2catid = {v: k for k, v in dataset.catid2clsid.items()}
-        anno_file = dataset.get_anno()
-        metric = COCOMetric(
-            anno_file=anno_file, clsid2catid=clsid2catid, IouType='bbox')
-    elif reader_cfg['metric'] == 'VOC':
-        metric = VOCMetric(
-            label_list=dataset.get_label_list(),
-            class_num=reader_cfg['num_classes'],
-            map_type=reader_cfg['map_type'])
-    else:
-        raise ValueError("metric currently only supports COCO and VOC.")
-    global_config['metric'] = metric
-
-    if 'Evaluation' in global_config.keys() and global_config['Evaluation']:
+    if 'Evaluation' in global_config.keys() and global_config[
+            'Evaluation'] and paddle.distributed.get_rank() == 0:
         eval_func = eval_function
+        dataset = reader_cfg['EvalDataset']
+        global val_loader
+        _eval_batch_sampler = paddle.io.BatchSampler(
+            dataset, batch_size=reader_cfg['EvalReader']['batch_size'])
+        val_loader = create('EvalReader')(dataset,
+                                          reader_cfg['worker_num'],
+                                          batch_sampler=_eval_batch_sampler,
+                                          return_list=True)
+        metric = None
+        if reader_cfg['metric'] == 'COCO':
+            clsid2catid = {v: k for k, v in dataset.catid2clsid.items()}
+            anno_file = dataset.get_anno()
+            metric = COCOMetric(
+                anno_file=anno_file, clsid2catid=clsid2catid, IouType='bbox')
+        elif reader_cfg['metric'] == 'VOC':
+            metric = VOCMetric(
+                label_list=dataset.get_label_list(),
+                class_num=reader_cfg['num_classes'],
+                map_type=reader_cfg['map_type'])
+        else:
+            raise ValueError("metric currently only supports COCO and VOC.")
+        global_config['metric'] = metric
     else:
         eval_func = None