search

princeton-vl / SEA-RAFT

[ECCV2024 - Oral, Best Paper Award Candidate] SEA-RAFT: Simple, Efficient, Accurate RAFT for Optical Flow
BSD 3-Clause "New" or "Revised" License
327 stars 11 forks source link

MoL loss for other models #3

Closed everRoc closed 5 months ago

everRoc commented 5 months ago

Great work! I wonder whether MoL loss is helpful for other methods, but I don't find the ablations in the paper. Have you ever directly adopted MoL loss on other models like RAFT and GMA?

MemorySlices commented 5 months ago

Thanks for the interest! Our work is very similar to RAFT and we have provided ablations. Unfortunately, we did not adopt MoL on other models such as GMA. According to our experience, the training settings of MoL and L1 for the same model are almost the same :D. You can have a quick ablation by replacing L1 with MoL according to the details mentioned in the paper.