Usage for 'padding_mask' in COCOInstanceNewBaselineDatasetMapper?
wenhe-jia opened this issue · 1 comments
wenhe-jia commented
Thanks to your excellent job!
I don't understand the usage of padding_mask in COCOInstanceNewBaselineDatasetMapper, it seems that it is no longer used in the following training process.
Can you explain this for me please? Thanks a lot.
bowenc0221 commented
The COCOInstanceNewBaselineDatasetMapper will always return an image with the same size (e.g., 1024x1024) with zero padding. padding_mask indicates which pixels are padding, which is not used yet.