Fix box cam error (#601)

* fix box cam

* fix box cam

configs/yolov5/yolov5_s-v61_fast_1xb12-40e_608x352_cat.py

@@ -0,0 +1,70 @@
+_base_ = 'yolov5_s-v61_fast_1xb12-40e_cat.py'
+
+# This configuration is used to provide non-square training examples
+# Must be a multiple of 32
+img_scale = (608, 352)  # w h
+
+anchors = [
+    [(65, 35), (159, 45), (119, 80)],  # P3/8
+    [(215, 77), (224, 116), (170, 166)],  # P4/16
+    [(376, 108), (339, 176), (483, 190)]  # P5/32
+]
+
+# ===============================Unmodified in most cases====================
+_base_.model.bbox_head.loss_obj.loss_weight = 1.0 * ((img_scale[1] / 640)**2)
+_base_.model.bbox_head.prior_generator.base_sizes = anchors
+
+train_pipeline = [
+    *_base_.pre_transform,
+    dict(
+        type='Mosaic',
+        img_scale=img_scale,
+        pad_val=114.0,
+        pre_transform=_base_.pre_transform),
+    dict(
+        type='YOLOv5RandomAffine',
+        max_rotate_degree=0.0,
+        max_shear_degree=0.0,
+        scaling_ratio_range=(1 - _base_.affine_scale, 1 + _base_.affine_scale),
+        # img_scale is (width, height)
+        border=(-img_scale[0] // 2, -img_scale[1] // 2),
+        border_val=(114, 114, 114)),
+    dict(
+        type='mmdet.Albu',
+        transforms=_base_.albu_train_transforms,
+        bbox_params=dict(
+            type='BboxParams',
+            format='pascal_voc',
+            label_fields=['gt_bboxes_labels', 'gt_ignore_flags']),
+        keymap={
+            'img': 'image',
+            'gt_bboxes': 'bboxes'
+        }),
+    dict(type='YOLOv5HSVRandomAug'),
+    dict(type='mmdet.RandomFlip', prob=0.5),
+    dict(
+        type='mmdet.PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape', 'flip',
+                   'flip_direction'))
+]
+
+_base_.train_dataloader.dataset.pipeline = train_pipeline
+
+test_pipeline = [
+    dict(type='LoadImageFromFile', file_client_args=_base_.file_client_args),
+    dict(type='YOLOv5KeepRatioResize', scale=img_scale),
+    dict(
+        type='LetterResize',
+        scale=img_scale,
+        allow_scale_up=False,
+        pad_val=dict(img=114)),
+    dict(type='mmdet.LoadAnnotations', with_bbox=True),
+    dict(
+        type='mmdet.PackDetInputs',
+        meta_keys=('img_id', 'img_path', 'ori_shape', 'img_shape',
+                   'scale_factor', 'pad_param'))
+]
+
+val_dataloader = dict(
+    dataset=dict(pipeline=test_pipeline, batch_shapes_cfg=None))
+test_dataloader = val_dataloader

demo/boxam_vis_demo.py

@@ -46,6 +46,9 @@ IGNORE_LOSS_PARAMS = {
     'yolov6': ['loss_cls'],
     'yolox': ['loss_obj'],
     'rtmdet': ['loss_cls'],
+    'yolov7': ['loss_obj'],
+    'yolov8': ['loss_cls'],
+    'ppyoloe': ['loss_cls'],
 }
 
 # This parameter is required in some algorithms

docs/zh_cn/tutorials/faq.md

@@ -105,3 +105,7 @@ unset PYTHONPATH
 7. `coco/bbox_mAP_l` 将会根据 `bbox_mAP_l` 作为筛选条件。
 
 此外用户还可以选择筛选的逻辑，通过设置配置中的 `default_hooks.checkpoint.rule` 来选择判断逻辑，如：`default_hooks.checkpoint.rule=greater` 表示指标越大越好。更详细的使用可以参考 [checkpoint_hook](https://github.com/open-mmlab/mmengine/blob/main/mmengine/hooks/checkpoint_hook.py) 来修改
+
+## 如何进行非正方形输入尺寸训练和测试 ?
+
+在 YOLO 系列算法中默认配置基本上都是 640x640 或者 1280x1280 正方形尺度输入训练的。用户如果想进行非正方形尺度训练，你可以修改配置中 `image_scale` 参数，并将其他对应位置进行修改即可。用户可以参考我们提供的 [yolov5_s-v61_fast_1xb12-40e_608x352_cat.py](https://github.com/open-mmlab/mmyolo/tree/dev/configs/yolov5/yolov5_s-v61_fast_1xb12-40e_608x352_cat.py) 配置。

mmyolo/utils/boxam_utils.py

@@ -203,6 +203,7 @@ class BoxAMDetectorWrapper(nn.Module):
             # Maybe this is a direction that can be optimized
             # self.detector.init_weights()
 
+            self.detector.bbox_head.head_module.training = True
             if hasattr(self.detector.bbox_head, 'featmap_sizes'):
                 # Prevent the model algorithm error when calculating loss
                 self.detector.bbox_head.featmap_sizes = None
@@ -218,6 +219,7 @@ class BoxAMDetectorWrapper(nn.Module):
 
             return [loss]
         else:
+            self.detector.bbox_head.head_module.training = False
             with torch.no_grad():
                 results = self.detector.test_step(self.input_data)
                 return results