[Fix] enable CocoMetric to use bbox score in pred_instances (open-mml…

…ab#2145)

demo/docs/2d_human_pose_demo.md

@@ -93,7 +93,7 @@ python demo/topdown_demo_with_mmdet.py \
     demo/mmdetection_cfg/faster_rcnn_r50_fpn_coco.py \
     https://download.openmmlab.com/mmdetection/v2.0/faster_rcnn/faster_rcnn_r50_fpn_1x_coco/faster_rcnn_r50_fpn_1x_coco_20200130-047c8118.pth \
     configs/body_2d_keypoint/topdown_heatmap/coco/td-hm_hrnet-w32_8xb64-210e_coco-256x192.py \
-    https://download.openmmlab.com/mmpose/top_down/hrnet/hrnet_w32_coco_256x192-c78dce93_20200708.pth \
+    https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/topdown_heatmap/coco/td-hm_hrnet-w32_8xb64-210e_coco-256x192-81c58e40_20220909.pth \
     --input tests/data/posetrack18/videos/000001_mpiinew_test/000001_mpiinew_test.mp4 \
     --output-root=vis_results/demo --show --draw-heatmap
 ```
@@ -116,7 +116,7 @@ Example:
 ```shell
 python demo/bottomup_demo.py \
     configs/body_2d_keypoint/dekr/coco/dekr_hrnet-w32_8xb10-140e_coco-512x512.py \
-    https://download.openmmlab.com/mmpose/bottom_up/dekr/hrnet_w32_coco_512x512-2a3056de_20220928.pth \
+    https://download.openmmlab.com/mmpose/v1/body_2d_keypoint/dekr/coco/dekr_hrnet-w32_8xb10-140e_coco-512x512_ac7c17bf-20221228.pth \
     --input tests/data/coco/000000197388.jpg --output-root=vis_results \
     --show --save-predictions
 ```

mmpose/apis/inferencers/mmpose_inferencer.py

@@ -176,6 +176,8 @@ def __call__(
         inputs = self.preprocess(
             inputs, batch_size=batch_size, **preprocess_kwargs)
 
+        forward_kwargs['bbox_thr'] = preprocess_kwargs.get('bbox_thr', -1)
+
         preds = []
         if 'pose2d' not in self.mode or not hasattr(self.pose2d_inferencer,
                                                     'detector'):

mmpose/apis/inferencers/pose2d_inferencer.py

@@ -184,10 +184,15 @@ def preprocess_single(self,
 
         return data_infos
 
-    def forward(self, inputs: Union[dict, tuple]):
+    def forward(self, inputs: Union[dict, tuple], bbox_thr=-1):
         data_samples = super().forward(inputs)
         if self.cfg.data_mode == 'topdown':
             data_samples = [merge_data_samples(data_samples)]
+        if bbox_thr > 0:
+            for ds in data_samples:
+                if 'bbox_scores' in ds.pred_instances:
+                    ds.pred_instances = ds.pred_instances[
+                        ds.pred_instances.bbox_scores > bbox_thr]
         return data_samples
 
     def __call__(
@@ -241,6 +246,7 @@ def __call__(
         else:
             inputs = self._inputs_to_list(inputs)
 
+        forward_kwargs['bbox_thr'] = preprocess_kwargs.get('bbox_thr', -1)
         inputs = self.preprocess(
             inputs, batch_size=batch_size, **preprocess_kwargs)
 

mmpose/evaluation/metrics/coco_metric.py

@@ -178,13 +178,19 @@ def process(self, data_batch: Sequence[dict],
             pred['keypoint_scores'] = keypoint_scores
             pred['category_id'] = data_sample.get('category_id', 1)
 
-            if ('bbox_scores' not in data_sample['gt_instances']
-                    or len(data_sample['gt_instances']['bbox_scores']) !=
-                    len(keypoints)):
+            if 'bbox_scores' in data_sample['pred_instances']:
+                # some one-stage models will predict bboxes and scores
+                # together with keypoints
+                bbox_scores = data_sample['pred_instances']['bbox_scores']
+            elif ('bbox_scores' not in data_sample['gt_instances']
+                  or len(data_sample['gt_instances']['bbox_scores']) !=
+                  len(keypoints)):
                 # bottom-up models might output different number of
                 # instances from annotation
                 bbox_scores = np.ones(len(keypoints))
             else:
+                # top-down models use detected bboxes, the scores of which
+                # are contained in the gt_instances
                 bbox_scores = data_sample['gt_instances']['bbox_scores']
             pred['bbox_scores'] = bbox_scores
 

mmpose/visualization/local_visualizer.py

@@ -264,12 +264,9 @@ def _draw_instances_kpts(self,
                 # compute neck joint
                 neck = np.mean(keypoints_info[:, [5, 6]], axis=1)
                 # neck score when visualizing pred
-                neck[:, 2] = 1 - np.logical_or(
-                    keypoints_info[:, 5, 2] < kpt_thr,
-                    keypoints_info[:, 6, 2] < kpt_thr).astype(int)
-                neck[:, 3] = 1 - np.logical_or(
-                    keypoints_info[:, 5, 3] < kpt_thr,
-                    keypoints_info[:, 6, 3] < kpt_thr).astype(int)
+                neck[:, 2:4] = np.logical_and(
+                    keypoints_info[:, 5, 2:4] < kpt_thr,
+                    keypoints_info[:, 6, 2:4] < kpt_thr).astype(int)
                 new_keypoints_info = np.insert(
                     keypoints_info, 17, neck, axis=1)
 

projects/yolox-pose/README.md

@@ -91,10 +91,10 @@ Results on COCO val2017
 
 |                              Model                              | Input Size |  AP   | AP<sup>50</sup> | AP<sup>75</sup> |  AR   | AR<sup>50</sup> |                                 Download                                 |
 | :-------------------------------------------------------------: | :--------: | :---: | :-------------: | :-------------: | :---: | :-------------: | :----------------------------------------------------------------------: |
-| [YOLOX-tiny-Pose](./configs/yolox-pose_tiny_4xb64-300e_coco.py) |    640     | 0.477 |      0.756      |      0.506      | 0.547 |      0.802      | [model](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_tiny_4xb64-300e_coco-c47dd83b_20230321.pth) \| [log](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_tiny_4xb64-300e_coco_20230321.json) |
-|    [YOLOX-s-Pose](./configs/yolox-pose_s_8xb32-300e_coco.py)    |    640     | 0.595 |      0.836      |      0.653      | 0.658 |      0.878      | [model](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_s_8xb32-300e_coco-9f5e3924_20230321.pth) \| [log](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_s_8xb32-300e_coco_20230321.json) |
-|    [YOLOX-m-Pose](./configs/yolox-pose_m_4xb64-300e_coco.py)    |    640     | 0.659 |      0.870      |      0.729      | 0.713 |      0.903      | [model](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_m_4xb64-300e_coco-cbd11d30_20230321.pth) \| [log](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_m_4xb64-300e_coco_20230321.json) |
-|    [YOLOX-l-Pose](./configs/yolox-pose_l_4xb64-300e_coco.py)    |    640     | 0.679 |      0.882      |      0.749      | 0.733 |      0.911      | [model](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_l_4xb64-300e_coco-122e4cf8_20230321.pth) \| [log](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_l_4xb64-300e_coco_20230321.json) |
+| [YOLOX-tiny-Pose](./configs/yolox-pose_tiny_4xb64-300e_coco.py) |    640     | 0.518 |      0.799      |      0.545      | 0.566 |      0.841      | [model](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_tiny_4xb64-300e_coco-c47dd83b_20230321.pth) \| [log](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_tiny_4xb64-300e_coco_20230321.json) |
+|    [YOLOX-s-Pose](./configs/yolox-pose_s_8xb32-300e_coco.py)    |    640     | 0.632 |      0.875      |      0.692      | 0.676 |      0.907      | [model](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_s_8xb32-300e_coco-9f5e3924_20230321.pth) \| [log](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_s_8xb32-300e_coco_20230321.json) |
+|    [YOLOX-m-Pose](./configs/yolox-pose_m_4xb64-300e_coco.py)    |    640     | 0.685 |      0.897      |      0.753      | 0.727 |      0.925      | [model](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_m_4xb64-300e_coco-cbd11d30_20230321.pth) \| [log](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_m_4xb64-300e_coco_20230321.json) |
+|    [YOLOX-l-Pose](./configs/yolox-pose_l_4xb64-300e_coco.py)    |    640     | 0.706 |      0.907      |      0.775      | 0.747 |      0.934      | [model](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_l_4xb64-300e_coco-122e4cf8_20230321.pth) \| [log](https://download.openmmlab.com/mmpose/v1/projects/yolox-pose/yolox-pose_l_4xb64-300e_coco_20230321.json) |
 
 We have only trained models with an input size of 640, as we couldn't replicate the performance enhancement mentioned in the paper when increasing the input size from 640 to 960. We warmly welcome any contributions if you can successfully reproduce the results from the paper!
 

projects/yolox-pose/configs/yolox-pose_s_8xb32-300e_coco.py

@@ -83,7 +83,7 @@
     test_cfg=dict(
         yolox_style=True,
         multi_label=False,
-        score_thr=0.2,
+        score_thr=0.001,
         max_per_img=300,
         nms=dict(type='nms', iou_threshold=0.65)))
 
@@ -189,7 +189,7 @@
 val_evaluator = dict(
     type='mmpose.CocoMetric',
     ann_file=data_root + 'annotations/person_keypoints_val2017.json',
-)
+    score_mode='bbox')
 test_evaluator = val_evaluator
 
 default_hooks = dict(checkpoint=dict(save_best='coco/AP', rule='greater'))