search

hitachinsk / FGT

[ECCV 2022] Flow-Guided Transformer for Video Inpainting
https://hitachinsk.github.io/publication/2022-10-01-Flow-Guided-Transformer-for-Video-Inpainting
MIT License
300 stars 31 forks source link

About square mask settings #36

Closed jinsuyoo closed 1 year ago

jinsuyoo commented 1 year ago

Hope everything is well with you!

Recently I've been trying to reproduce FGT's performance on square mask settings but failed. Under our own setting (by using "object_removal.yaml" config), we face that the FGT's performance on Youtube-VOS is lower than STTN about 1dB, while outperforming STTN on the DAVIS dataset. Can you share the config for square settings if exist?

Thank you for open-sourcing your great work :)

Best,

Jinsu

hitachinsk commented 1 year ago

In square mask setting, we adopt the waterwark_removal.yaml as the default config. Please use this config to reproduce the results.