search

Huangying-Zhan / DF-VO

Depth and Flow for Visual Odometry
https://df-vo.readthedocs.io
MIT License
565 stars 117 forks source link

No modification, but strange results #13

Closed Nyohohoho closed 4 years ago

Nyohohoho commented 4 years ago

Thank you for your excellent work and kind sharing.

I tried to run your code. I downloaded the code and pretrained models without any modification. I only modified the path so that KITTI dataset on my machine can be accessed.

However, I got this kind of result... Screenshot 2020-09-24 201202

I cannot figure out the reason.

DanRos22 commented 4 years ago

It seems you face the same problem I had. If you see a warning regarding torch.nn.functional.grid_sample function, you need to add the flag align_corners = True (4 times across this project), in order to get the original behavior. This change is needed apparently because you use a newer pytorch version compared to this repo requirements (1.1.0), and according to pytorch documentation: The default behavior up to version 1.2.0 was ``align_corners = True``. Since then, the default behavior has been changed to ``align_corners = False``, @Huangying-Zhan, Maybe you can update your code so any new user won't experience the same difficulty. Thanks! And BTW, thanks for sharing this work. it's great!

Huangying-Zhan commented 4 years ago

@DanRos22 Thanks for pointing out the issue and the suggestion.