search

yoxu515 / aot-benchmark

An efficient modular implementation of Associating Objects with Transformers for Video Object Segmentation in PyTorch
BSD 3-Clause "New" or "Revised" License
600 stars 108 forks source link

F score over the boundaries #34

Closed bhack closed 1 year ago

bhack commented 1 year ago

It could be nice to have the F score for the boundaries other the IOU in utils/metrics.

z-x-yang commented 1 year ago

Thanks. A good suggestion, we'll work on this.

bhack commented 1 year ago

If you are interested I have replicated boundaries extraction reference implementation of BoundaryIOU in Pytorch:

https://github.com/Lightning-AI/metrics/issues/1500#issuecomment-1433162536

z-x-yang commented 1 year ago

Thank you for your suggestion. I have reviewed the implementation method in the link you provided, and while it is indeed a good approach, it cannot be directly applied to the AOT code. This is because the calculation of IOU in the aot-benchmark uses multi-object masks rather than binary masks. However, I will implement a multi-target version in recent weeks.

bhack commented 1 year ago

That one was only the mask to boundaries then In the official boundary IOU it was just looped over annotations masks:

 https://github.com/bowenc0221/boundary-iou-api/blob/master/boundary_iou/utils/boundary_utils.py#L37-L38