[Bugfix] test resize with pad_shape (open-mmlab#3421)

## Motivation

When using `test_cfg` for `data_preprocessor`, `predict_by_feat` resizes
to the original size, not the padded size.

```
data_preprocessor = dict(
    type="SegDataPreProcessor",
    #type="SegDataPreProcessorWithPad",
    mean=[123.675, 116.28, 103.53],
    std=[58.395, 57.12, 57.375],
    bgr_to_rgb=True,
    pad_val=0,
    seg_pad_val=255,
    test_cfg=dict(size=(128, 128)))
```

Refar to:

https://github.com/open-mmlab/mmsegmentation/blob/main/mmseg/models/decode_heads/san_head.py#L589-L592

## Checklist

1. Pre-commit or other linting tools are used to fix the potential lint
issues.
2. The modification is covered by complete unit tests. If not, please
add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects,
this PR should be tested with downstream projects, like MMDet or
MMDet3D.
4. The documentation has been modified accordingly, like docstring or
example tutorials.

mmseg/models/decode_heads/decode_head.py

@@ -350,9 +350,17 @@ def predict_by_feat(self, seg_logits: Tensor,
             Tensor: Outputs segmentation logits map.
         """
 
+        if isinstance(batch_img_metas[0]['img_shape'], torch.Size):
+            # slide inference
+            size = batch_img_metas[0]['img_shape']
+        elif 'pad_shape' in batch_img_metas[0]:
+            size = batch_img_metas[0]['pad_shape'][:2]
+        else:
+            size = batch_img_metas[0]['img_shape']
+
         seg_logits = resize(
             input=seg_logits,
-            size=batch_img_metas[0]['img_shape'],
+            size=size,
             mode='bilinear',
             align_corners=self.align_corners)
         return seg_logits

mmseg/models/decode_heads/san_head.py

@@ -586,8 +586,11 @@ def predict_by_feat(self, seg_logits: List[Tensor],
         """
         mask_pred = seg_logits[0]
         cls_score = seg_logits[1]
-        if 'pad_shape' in batch_img_metas[0]:
-            size = batch_img_metas[0]['pad_shape']
+        if isinstance(batch_img_metas[0]['img_shape'], torch.Size):
+            # slide inference
+            size = batch_img_metas[0]['img_shape']
+        elif 'pad_shape' in batch_img_metas[0]:
+            size = batch_img_metas[0]['pad_shape'][:2]
         else:
             size = batch_img_metas[0]['img_shape']
         # upsample mask

projects/hssn/decode_head/sep_aspp_contrast_head.py

@@ -127,10 +127,17 @@ def predict_by_feat(self, seg_logits: Tuple[Tensor],
         # elif seg_logit.size(1) == 144 # For Mapillary dataset, 124+16+4
         # unofficial repository not release mapillary until 2023/2/6
 
+        if isinstance(batch_img_metas[0]['img_shape'], torch.Size):
+            # slide inference
+            size = batch_img_metas[0]['img_shape']
+        elif 'pad_shape' in batch_img_metas[0]:
+            size = batch_img_metas[0]['pad_shape'][:2]
+        else:
+            size = batch_img_metas[0]['img_shape']
         seg_logit = seg_logit[:, :-hiera_num_classes]
         seg_logit = resize(
             input=seg_logit,
-            size=batch_img_metas[0]['img_shape'],
+            size=size,
             mode='bilinear',
             align_corners=self.align_corners)