[Fix] Fix the bug in binary cross entropy loss (#1499)

* [Fix] Fix the bug in binary cross entropy loss

 Fix the bug in binary cross entropy loss when using multi-label datasets e.g.VOC2007

* update ci

---------

Co-authored-by: Ezra-Yu <18586273+Ezra-Yu@users.noreply.github.com>

.github/workflows/pr_stage_test.yml

@@ -34,7 +34,7 @@ jobs:
       - name: Upgrade pip
         run: pip install pip --upgrade
       - name: Install PyTorch
-        run: pip install torch==${{matrix.torch}}+cpu torchvision==${{matrix.torchvision}}+cpu -f https://download.pytorch.org/whl/torch_stable.html
+        run: pip install torch==${{matrix.torch}}+cpu torchvision==${{matrix.torchvision}}+cpu -f https://download.pytorch.org/whl/cpu/torch_stable.html
       - name: Install mmpretrain dependencies
         run: |
           pip install git+https://github.com/open-mmlab/mmengine.git@main
@@ -42,7 +42,7 @@ jobs:
           mim install 'mmcv >= 2.0.0rc4'
           pip install -r requirements.txt
       - name: Build and install
-        run: pip install -e .
+        run: mim install .
       - name: Run unittests and generate coverage report
         run: |
           coverage run --branch --source mmpretrain -m pytest tests/
@@ -129,7 +129,7 @@ jobs:
           mim install 'mmcv >= 2.0.0rc4'
           pip install -r requirements.txt
       - name: Build and install
-        run: pip install -e . -v
+        run: mim install .
       - name: Run unittests
         run: |
           pytest tests/ --ignore tests/test_models/test_backbones

.github/workflows/test_mim.yml

@@ -39,6 +39,6 @@ jobs:
       - name: Install openmim
         run: pip install openmim
       - name: Build and install
-        run: mim install -e .
+        run: mim install .
       - name: test commands of mim
         run: mim search mmpretrain

mmpretrain/models/heads/multi_label_cls_head.py

@@ -93,7 +93,7 @@ class MultiLabelClsHead(BaseModule):
         num_classes = cls_score.size()[-1]
         # Unpack data samples and pack targets
         if 'gt_score' in data_samples[0]:
-            target = torch.stack([i.gt_score for i in data_samples])
+            target = torch.stack([i.gt_score.float() for i in data_samples])
         else:
             target = torch.stack([
                 label_to_onehot(i.gt_label, num_classes) for i in data_samples