Dev 1.x Adapt PETR to mmcv v2 version dataset (#2800)

projects/PETR/README.md

@@ -16,7 +16,7 @@ This is an implementation of *PETR*.
 In MMDet3D's root directory, run the following command to train the model:
 
 ```bash
-python tools/train.py projects/PETR/config/petr/petr_vovnet_gridmask_p4_800x320.py
+python tools/train.py projects/PETR/configs/petr_vovnet_gridmask_p4_800x320.py
 ```
 
 ### Testing commands
@@ -24,7 +24,7 @@ python tools/train.py projects/PETR/config/petr/petr_vovnet_gridmask_p4_800x320.
 In MMDet3D's root directory, run the following command to test the model:
 
 ```bash
-python tools/test.py projects/PETR/config/petr/petr_vovnet_gridmask_p4_800x320.py ${CHECKPOINT_PATH}
+python tools/test.py projects/PETR/configs/petr_vovnet_gridmask_p4_800x320.py ${CHECKPOINT_PATH}
 ```
 
 ## Results

projects/PETR/petr/petr_head.py

@@ -446,7 +446,7 @@ class PETRHead(AnchorFreeHead):
         masks = x.new_ones((batch_size, num_cams, input_img_h, input_img_w))
         for img_id in range(batch_size):
             for cam_id in range(num_cams):
-                img_h, img_w, _ = img_metas[img_id]['img_shape'][cam_id]
+                img_h, img_w = img_metas[img_id]['img_shape'][cam_id]
                 masks[img_id, cam_id, :img_h, :img_w] = 0
         x = self.input_proj(x.flatten(0, 1))
         x = x.view(batch_size, num_cams, *x.shape[-3:])