Merge branch 'meituan:main' into patch-1

deploy/OpenVINO/README.md

@@ -15,4 +15,10 @@ python deploy/OpenVINO/export_openvino.py --weights yolov6s.pt --img 640 --batch
 ### Download
 * [YOLOv6-nano](https://github.com/meituan/YOLOv6/releases/download/0.1.0/yolov6n_openvino.tar.gz)
 * [YOLOv6-tiny](https://github.com/meituan/YOLOv6/releases/download/0.1.0/yolov6n_openvino.tar.gz)
-* [YOLOv6-s](https://github.com/meituan/YOLOv6/releases/download/0.1.0/yolov6n_openvino.tar.gz)
+* [YOLOv6-s](https://github.com/meituan/YOLOv6/releases/download/0.1.0/yolov6n_openvino.tar.gz)
+
+### Speed test
+```shell
+benchmark_app -m yolov6s_openvino/yolov6s.xml -i data/images/image1.jpg -d CPU -niter 100 -progress 
+
+```

docs/Test_speed.md

@@ -1,6 +1,6 @@
 # Test speed
 
-This guidence explains how to reproduce speed results of YOLOv6. For fair comparision, the speed results do not contain the time cost of data pre-processing and NMS post-processing.
+This guidence explains how to reproduce speed results of YOLOv6. For fair comparison, the speed results do not contain the time cost of data pre-processing and NMS post-processing.
 
 ## 0. Prepare model
 

docs/Train_custom_data.md

@@ -1,6 +1,6 @@
 # Train Custom Data
 
-This guidence explains how to train your own custom data with YOLOv6 ( take fine-tuning YOLOv6-s model for example).
+This guidence explains how to train your own custom data with YOLOv6 (take fine-tuning YOLOv6-s model for example).
 
 ## 0. Before you start
 

tools/eval.py

@@ -46,7 +46,7 @@ def run(data,
         ):
     """ Run the evaluation process
 
-    This function is the main process of evalutaion, supporting image file and dir containing images.
+    This function is the main process of evaluataion, supporting image file and dir containing images.
     It has tasks of 'val', 'train' and 'speed'. Task 'train' processes the evaluation during training phase.
     Task 'val' processes the evaluation purely and return the mAP of model.pt. Task 'speed' precesses the
     evaluation of inference speed of model.pt.

tools/train.py

@@ -26,7 +26,10 @@ def get_args_parser(add_help=True):
     parser.add_argument('--epochs', default=400, type=int, help='number of total epochs to run')
     parser.add_argument('--workers', default=8, type=int, help='number of data loading workers (default: 8)')
     parser.add_argument('--device', default='0', type=str, help='cuda device, i.e. 0 or 0,1,2,3 or cpu')
-    parser.add_argument('--noval', action='store_true', help='only evaluate in final epoch')
+    parser.add_argument('--eval-interval', default=20, help='evaluate at every interval epochs')
+    parser.add_argument('--eval-final-only', action='store_true', help='only evaluate at the final epoch')
+    parser.add_argument('--heavy-eval-range', default=50,
+                        help='evaluating every epoch for last such epochs (can be jointly used with --eval-interval)')
     parser.add_argument('--check-images', action='store_true', help='check images when initializing datasets')
     parser.add_argument('--check-labels', action='store_true', help='check label files when initializing datasets')
     parser.add_argument('--output-dir', default='./runs/train', type=str, help='path to save outputs')

yolov6/core/engine.py

@@ -98,9 +98,9 @@ def train_in_steps(self):
         self.update_optimizer()
 
     def eval_and_save(self):
-        epoch_sub = self.max_epoch - self.epoch
-        val_period = 20 if epoch_sub > 100 else 1 # to fasten training time, evaluate in every 20 epochs for the early stage.
-        is_val_epoch = (not self.args.noval or (epoch_sub == 1)) and (self.epoch % val_period == 0)
+        remaining_epochs = self.max_epoch - self.epoch
+        eval_interval = self.args.eval_interval if remaining_epochs > self.args.heavy_eval_range else 1
+        is_val_epoch = (not self.args.eval_final_only or (remaining_epochs == 1)) and (self.epoch % eval_interval == 0)
         if self.main_process:
             self.ema.update_attr(self.model, include=['nc', 'names', 'stride']) # update attributes for ema model
             if is_val_epoch:

yolov6/core/evaler.py

@@ -105,7 +105,7 @@ def predict_model(self, model, dataloader, task):
     def eval_model(self, pred_results, model, dataloader, task):
         '''Evaluate models
         For task speed, this function only evaluates the speed of model and outputs inference time.
-        For task val, this function evalutates the speed and mAP by pycocotools, and returns 
+        For task val, this function evaluates the speed and mAP by pycocotools, and returns 
         inference time and mAP value.
         '''
         LOGGER.info(f'\nEvaluating speed.')

yolov6/layers/common.py

@@ -139,7 +139,7 @@ class RepVGGBlock(nn.Module):
     def __init__(self, in_channels, out_channels, kernel_size=3,
                  stride=1, padding=1, dilation=1, groups=1, padding_mode='zeros', deploy=False, use_se=False):
         super(RepVGGBlock, self).__init__()
-        """ Intialization of the class.
+        """ Initialization of the class.
         Args:
             in_channels (int): Number of channels in the input image
             out_channels (int): Number of channels produced by the convolution

yolov6/utils/figure_iou.py

@@ -13,7 +13,7 @@ def __init__(self, box_format='xywh', iou_type='ciou', reduction='none', eps=1e-
             box_format: (string), must be one of 'xywh' or 'xyxy'.
             iou_type: (string), can be one of 'ciou', 'diou', 'giou' or 'siou'
             reduction: (string), specifies the reduction to apply to the output, must be one of 'none', 'mean','sum'.
-            eps: (float), a value to avoid devide by zero error.
+            eps: (float), a value to avoid divide by zero error.
         """
         self.box_format = box_format
         self.iou_type = iou_type.lower()