search

gangweiX / Fast-ACVNet

[TPAMI 2024] Fast-ACV: Fast Attention Concatenation Volume for Accurate and Real-time Stereo Matching
MIT License
356 stars 37 forks source link

Fast_ACVNet and Fast_ACVNet_plus #6

Closed Torment123 closed 1 year ago

Torment123 commented 1 year ago

Hi, I noticed that there's a fast_acvnet_plus in the model folder, which is implemented differently compare to the paper (the correlation is computed between feature maps of 1/4 of the original resolution instead of 1/8), So if I want to reproduce the result of fast_acvnet, I think I should instead use the model in Fast_ACV.py ? Also, what benefit of fast_acvnet_plus compare to the vanilla fast_acvnet? Thanks in davance

gangweiX commented 1 year ago

Thank you for your attention. The fast_acvnet_plus constructs a higher resolution correlation (i.e. 1/4), which can obtain more accurate attention weights.

Torment123 commented 1 year ago

I see, thanks for your fast reponse