[Fix] Fix Mask2Former (#8436)

* fix typos * fix detr * remove redundant code * update * remove redundant code
open-mmlab · Aug 17, 2022 · e799c4c · e799c4c
1 parent b564ad3
commit e799c4c
Show file tree

Hide file tree

Showing 2 changed files with 9 additions and 7 deletions.
diff --git a/mmdet/models/dense_heads/detr_head.py b/mmdet/models/dense_heads/detr_head.py
@@ -485,9 +485,14 @@ def _get_targets_single(self, cls_score: Tensor, bbox_pred: Tensor,
             - pos_inds (Tensor): Sampled positive indices for each image.
             - neg_inds (Tensor): Sampled negative indices for each image.
         """
-
+        img_h, img_w = img_meta['img_shape']
+        factor = bbox_pred.new_tensor([img_w, img_h, img_w,
+                                       img_h]).unsqueeze(0)
         num_bboxes = bbox_pred.size(0)
+        # convert bbox_pred from xywh, normalized to xyxy, unnormalized
         bbox_pred = bbox_cxcywh_to_xyxy(bbox_pred)
+        bbox_pred = bbox_pred * factor
+
         pred_instances = InstanceData(scores=cls_score, bboxes=bbox_pred)
         # assigner and sampler
         assign_result = self.assigner.assign(
@@ -515,13 +520,10 @@ def _get_targets_single(self, cls_score: Tensor, bbox_pred: Tensor,
         bbox_targets = torch.zeros_like(bbox_pred)
         bbox_weights = torch.zeros_like(bbox_pred)
         bbox_weights[pos_inds] = 1.0
-        img_h, img_w = img_meta['img_shape']
 
         # DETR regress the relative position of boxes (cxcywh) in the image.
         # Thus the learning target should be normalized by the image size, also
         # the box format should be converted from defaultly x1y1x2y2 to cxcywh.
-        factor = bbox_pred.new_tensor([img_w, img_h, img_w,
-                                       img_h]).unsqueeze(0)
         pos_gt_bboxes_normalized = pos_gt_bboxes / factor
         pos_gt_bboxes_targets = bbox_xyxy_to_cxcywh(pos_gt_bboxes_normalized)
         bbox_targets[pos_inds] = pos_gt_bboxes_targets

diff --git a/mmdet/models/dense_heads/mask2former_head.py b/mmdet/models/dense_heads/mask2former_head.py
@@ -169,9 +169,9 @@ def init_weights(self) -> None:
             if p.dim() > 1:
                 nn.init.xavier_normal_(p)
 
-    def _get_target_single(self, cls_score: Tensor, mask_pred: Tensor,
-                           gt_instances: InstanceData,
-                           img_meta: dict) -> Tuple[Tensor]:
+    def _get_targets_single(self, cls_score: Tensor, mask_pred: Tensor,
+                            gt_instances: InstanceData,
+                            img_meta: dict) -> Tuple[Tensor]:
         """Compute classification and mask targets for one image.
 
         Args: