diff --git a/.github/workflows/build.yml b/.github/workflows/build.yml
index a454e0664c..83fd2d219d 100644
--- a/.github/workflows/build.yml
+++ b/.github/workflows/build.yml
@@ -82,7 +82,7 @@ jobs:
         run: pip install torch==${{ matrix.torch }} torchvision==${{ matrix.torchvision }} -f https://download.pytorch.org/whl/torch_stable.html
       - name: Install MMCV ${{ matrix.mmcv_link }}
         run: |
-          pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/${{ matrix.mmcv_link }}/index.html --only-binary mmcv-full "opencv-python<=4.5.4.60"
+          pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/${{ matrix.mmcv_link }}/index.html --only-binary mmcv-full
           python -c 'import mmcv; print(mmcv.__version__)'
       - name: Build and install
         run: pip install -e .[all]
diff --git a/requirements/runtime.txt b/requirements/runtime.txt
index d2ddfa08c7..07be58f199 100644
--- a/requirements/runtime.txt
+++ b/requirements/runtime.txt
@@ -4,13 +4,12 @@ facexlib
 lmdb
 mmcv-full>=1.3.13 # To support DCN on CPU
 numpy
-opencv-python<=4.5.4.60
+opencv-python!=4.5.5.62,!=4.5.5.64
 # MMCV depends opencv-python instead of headless, thus we install opencv-python
-# Due to a binary compatibility bug, we limit its version before 4.5.5
+# Due to a bug from upstream, we skip this two version
 # https://github.com/opencv/opencv-python/issues/602
 # https://github.com/opencv/opencv/issues/21366
 # It seems to be fixed in https://github.com/opencv/opencv/pull/21382
-# But our test is negative.
 Pillow
 tensorboard
 torch