search

haofeixu / gmflow

[CVPR'22 Oral] GMFlow: Learning Optical Flow via Global Matching
Apache License 2.0
633 stars 54 forks source link

Flow Propagation in GMFlow (CVPR22) vs Flow Field Deformation in LiteFlowNet3 (ECCV20) #7

Closed twhui closed 2 years ago

twhui commented 2 years ago

Congratulations on your new work!

I think the idea of flow propagation in GMFlow is similar to the flow field deformation in LiteFlowNet3. Both of us propagate flow predictions and measure feature self-similarity. However, LiteFlowNet3 was not cited in GMFlow.

haofeixu commented 2 years ago

Hi @twhui , thank you for bringing the flow field deformation in LiteFlowNet3 to our attention, sorry that we are not aware of this approach while working on our GMFlow project.

Actually, our flow propagation was inspired by the GMA (ICCV 2021) paper, as we have mentioned in our paper. However, I do agree that our high-level idea is similar after checking the LiteFlowNet3 paper despite some difference in implementations. Unfortunately the CVPR camera ready deadline has passed, but I will add LiteFlowNet3 in our extended paper in future. Thanks and wish all the best!

twhui commented 2 years ago

Hi Haofei, I think you can cite LiteFlowNet3 in your updated arxiv version (https://arxiv.org/abs/2111.13680). Do you think this is possible?

haofeixu commented 2 years ago

Hi @twhui ,

FYI, I have referenced LiteFlowNet3 in our updated arxiv paper.

twhui commented 2 years ago

Thank you.