refine PaddleDetection benchmard print (#1556)

* refine PaddleDetection benchmard print, test=develop

* refine PaddleDetection benchmard print, test=develop

* refine PaddleDetection benchmard print, test=develop

* refine PaddleDetection benchmard print, test=develop

configs/cascade_rcnn_r50_1x.yml

 architecture: CascadeRCNN
 use_gpu: true
 max_iters: 180000
-log_smooth_window: 50
+log_iter: 50
 save_dir: output
 snapshot_iter: 10000
 pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/dygraph/resnet50.pdparams

configs/faster_rcnn_r50_1x.yml

 architecture: FasterRCNN
 use_gpu: true
 max_iters: 180000
-log_smooth_window: 50
+log_iter: 50
 save_dir: output
 snapshot_iter: 10000
 pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/dygraph/resnet50.pdparams

configs/mask_rcnn_r50_1x.yml

 architecture: MaskRCNN
 use_gpu: true
 max_iters: 180000
-log_smooth_window: 50
+log_iter: 50
 save_dir: output
 snapshot_iter: 10000
 pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/dygraph/resnet50.pdparams

configs/mask_rcnn_r50_fpn_1x.yml

 architecture: MaskRCNN
 use_gpu: true
 max_iters: 180000
-log_smooth_window: 20
+log_iter: 20
 save_dir: output
 snapshot_iter: 10000
 pretrain_weights: https://paddle-imagenet-models-name.bj.bcebos.com/ResNet50_cos_pretrained.tar

configs/yolov3_darknet.yml

 architecture: YOLOv3
 use_gpu: true
 max_iters: 500000
-log_smooth_window: 20
+log_iter: 20
 save_dir: output
 snapshot_iter: 50000
 metric: COCO

tools/train.py

@@ -132,7 +132,7 @@ def run(FLAGS, cfg):
     train_reader = create_reader(
         cfg.TrainReader, (cfg.max_iters - start_iter), cfg, devices_num=1)
 
-    time_stat = deque(maxlen=cfg.log_smooth_window)
+    time_stat = deque(maxlen=cfg.log_iter)
     start_time = time.time()
     end_time = time.time()
     # Run Train 
@@ -167,13 +167,13 @@ def run(FLAGS, cfg):
         if ParallelEnv().nranks < 2 or ParallelEnv().local_rank == 0:
             # Log state 
             if iter_id == 0:
-                train_stats = TrainingStats(cfg.log_smooth_window,
-                                            outputs.keys())
+                train_stats = TrainingStats(cfg.log_iter, outputs.keys())
             train_stats.update(outputs)
             logs = train_stats.log()
             if iter_id % cfg.log_iter == 0:
-                strs = 'iter: {}, lr: {:.6f}, {}, time: {:.3f}, eta: {}'.format(
-                    iter_id, curr_lr, logs, time_cost, eta)
+                ips = float(cfg['TrainReader']['batch_size']) / time_cost
+                strs = 'iter: {}, lr: {:.6f}, {}, eta: {}, batch_cost: {:.5f} sec, ips: {:.5f} images/sec'.format(
+                    iter_id, curr_lr, logs, eta, time_cost, ips)
                 logger.info(strs)
             # Save Stage 
             if iter_id > 0 and iter_id % int(