Max Zhaoshuo Li 李赵硕
Max Zhaoshuo Li 李赵硕
Nice catch. It is worth updating the manuscript to further clarify this, which you are right that it will be stricter than just 3px error. The caption tries to state...
Yes ;) The arxiv/github results are outdated and I haven't synced them with the KITTI benchmark yet. The update will come soon (probably next week)
Hi @minchong1998, it's now updated.
Hi @Miaowei-HNU these results are from KITTI test data, calculated by KITTI website. ‘bg’ refers to background. ‘fg’ refers to foreground.
Hi @Miaowei-HNU , L1-raw is the metric of the cross-attention raw disparity at a lower resolution, which ideally should be low similarly to L1. In KITTI 2015 however, we have...
Hi @xzjzsa Sorry I didn't get your question. Can you elaborate? Training on KITTI and testing on KITTI or training on Scene Flow and testing on KITTI? - For training...
Hi @xzjzsa Good question and I have never tried it. However, this in theory wouldn't work very well because KITTI is a very very small dataset and there is just...
Hi @xzjzsa, My understanding is that you are training on SceneFlow and you are getting 7.75 EPE on SceneFlow? And you are saying that provided SceneFlow model gives you 16.82...
Hi @xzjzsa Oh I see what you mean now. Have you seen this issue #8? In short, torch version will lead to your problem. I assume you are not using...
Hmm, can you share the data with me? Maybe dump the images and disparity and share me a link so I can take a look