search

autonomousvision / unimatch

[TPAMI'23] Unifying Flow, Stereo and Depth Estimation
https://haofeixu.github.io/unimatch/
MIT License
980 stars 102 forks source link

Pretrained Model in Optical Flow Task #17

Closed RickyYXY closed 1 year ago

RickyYXY commented 1 year ago

I think when training unimatch begining at chairs(optical flow) dataset, we don't need a pretrained model. But in your gmflow_scale2_regrefine6_train.sh, you add "resume" in the script, and the resume target gmflow-scale2-chairs-020cc9be.pth can't be found in model_zoo. Can you explain why? Thks.

haofeixu commented 1 year ago

Sure the model can be trained from random initialization, but using a pretrained model can make the training converge faster. Sorry for missing the model gmflow-scale2-chairs-020cc9be.pth in our model zoo, it's available here: https://drive.google.com/file/d/1P3PrD636mQDQwMAK4Cv8ADoXUQ8QLMVD/view?usp=sharing

RickyYXY commented 1 year ago

OK, I get it! Thanks for your reply and code!