What are the meanings of "join label" and sample['meta']['flip']

z-x-yang / CFBI

The official implementation of CFBI(+): Collaborative Video Object Segmentation by (Multi-scale) Foreground-Background Integration.

BSD 3-Clause "New" or "Revised" License

322 stars 43 forks source link

What are the meanings of "join label" and sample['meta']['flip'] #29

Closed houhouhouhou11 closed 3 years ago

houhouhouhou11 commented 3 years ago

Hello: thanks for your sharing the perfect work! I have question in eval_manager.py which is the meanings of "join label" and sample['meta']['flip'].What are they represent? Thank you very much!

z-x-yang commented 3 years ago

1) "join label". Please refer to this issue.

2) "flip," used for testing time augmentation, flipping. Unable in default.

houhouhouhou11 commented 3 years ago

@z-x-yang sorry, I haven't understand clearly! Is "flip" used for testing time augmentation? what is the meaning of all_pred = flip_tensor(all_pred, 3) ? thank you

z-x-yang commented 3 years ago

When enabling flipping, the videos will be additionally evaluated by flipping each frame horizontally. Thus, we have to flip back the flipped predictions before ensembling both un-flipped and flipped results.

z-x-yang commented 3 years ago

The flipping ensemble has been commonly used in semantic segmentation. We just apply the same strategy to video object segmentation.