Closed NickHerrig closed 6 months ago
Never mind, should have read the paper.
Closing this
Never mind, should have read the paper.
Closing this
The previous link on kaggle for coconut-B only contains the extra split from unlabeled set and will need one more step to merge the files. But last week I already update the kaggle link to version 9 which contains the full set of coconut-B and no extra merging steps are needed. This updated version should contain 242k masks.
The COCONut-B image download link downloads a zip file with only 118k images instead of the documented 241k images.
I noticed that the provided annotations have 241690 masks, however when downloading the image link there are only 120k images.
I'm using the link in the provided markdown below:
Should I expect to see the full 242k Images in this image download link?
Or is the idea that COCONut-B = Train2017 + Unlabeled2017 + Val2017