polish the yolov3 dygraph (#4343)

* remove num_workers

* fix devices_num

* fix data loader

* remove args syncbn

* fix typo

dygraph/yolov3/README.md

@@ -139,7 +139,6 @@ Train Loss
 
 模型评估是指对训练完毕的模型评估各类性能指标。本示例采用[COCO官方评估](http://cocodataset.org/#detections-eval)
 
-    sh ./weights/download.sh
 
 `eval.py`是评估模块的主要执行程序，调用示例如下：
 
@@ -151,6 +150,8 @@ Train Loss
 - 通过设置`export CUDA_VISIBLE_DEVICES=0`指定单卡GPU评估。
 
 
+
+
 ## 进阶使用
 
 ### 背景介绍

dygraph/yolov3/reader.py

@@ -197,6 +197,7 @@ class DataSetReader(object):
                                             gt_scores, mixup_im, mixup_gt_boxes,
                                             mixup_gt_labels, mixup_gt_scores)
 
+
             im, gt_boxes, gt_labels, gt_scores = \
                 image_utils.image_augment(im, gt_boxes, gt_labels,
                                           gt_scores, size, mean)
@@ -326,9 +327,10 @@ def train(size=416,
         data_loader.set_sample_list_generator(infinite_reader,places=place)
         generator_out = []
         for data in data_loader(): 
-            for i in data:
-                generator_out.append(i.numpy()[0])              
-            yield [generator_out]
+            im, gt_boxes, gt_labels, gt_scores = data[0].numpy(),data[1].numpy(),data[2].numpy(),data[3].numpy()
+            for i in range(batch_size):
+                generator_out.append([im[i],gt_boxes[i],gt_labels[i],gt_scores[i]])
+            yield generator_out
             generator_out = []
             cnt += 1
             if cnt >= total_iter:

dygraph/yolov3/train.py

@@ -67,9 +67,10 @@ def train():
         os.makedirs(cfg.model_save_dir)
 
     gpu_id = int(os.environ.get('FLAGS_selected_gpus', 0))
+    if cfg.use_gpu:
         place = fluid.CUDAPlace(fluid.dygraph.parallel.Env().dev_id) if cfg.use_data_parallel else fluid.CUDAPlace(0)
-    if not cfg.use_gpu:
-        palce = fluid.CPUPlace()
+    else:
+        place = fluid.CPUPlace()
     with fluid.dygraph.guard(place):
         if args.use_data_parallel:
             strategy = fluid.dygraph.parallel.prepare_context()

dygraph/yolov3/utility.py

@@ -133,7 +133,6 @@ def parse_args():
     add_arg('no_mixup_iter',    int,    40000,  "Disable mixup in last N iter.")
     # TRAIN TEST INFER
     add_arg('input_size',       int,    608,    "Image input size of YOLOv3.")
-    add_arg('syncbn',           bool,   True,   "Whether to use synchronized batch normalization.")
     add_arg('random_shape',     bool,   True,   "Resize to random shape for train reader.")
     add_arg('valid_thresh',     float,  0.005,  "Valid confidence score for NMS.")
     add_arg('nms_thresh',       float,  0.45,   "NMS threshold.")