search

gangweiX / CGI-Stereo

A novel neural network architecture that can concurrently achieve real-time performance, competitive accuracy, and strong generalization ability.
MIT License
313 stars 29 forks source link

Confusion about sceneflow results of ACVNet and CGI-Stereo #1

Closed Torment123 closed 1 year ago

Torment123 commented 1 year ago

Hi, thanks for the very insightful work of newly developed CGI module. While reading the paper, I found that in Table 2 it shows that on sceneflow, the EPE of ACVNet is 0.64, while the reported result in the ACV paper is 0.48, could you help me clarify which version of ACVNet this table is referring to? Thanks.

gangweiX commented 1 year ago

I'm sorry for that. The results of 0.64 in Table 2 (CGI) is different from ACVNet. In Table 2, we only use ACV as the cost volume, and the other components of the network remain the same as CGI-Stereo. For example, we use the lighter mobilenet to extract features instead of Resnet, and we also use a more lightweight cost aggregation network, which is more lightweight than ACVNet.

Torment123 commented 1 year ago

Thanks for the clarification