search

ClementPinard / SfmLearner-Pytorch

Pytorch version of SfmLearner from Tinghui Zhou et al.
MIT License
1.01k stars 224 forks source link

depth images seem the same #119

Closed JIAJIAqi-1 closed 3 years ago

JIAJIAqi-1 commented 3 years ago

Hi! Thank you very much for your work. When I try to plot depth images, I found the results is exactly the same. How can I get the depth images in the paper? Thanks! 000000_depth 004540_depth

ClementPinard commented 3 years ago

Hi, what were the input images ? It is expected that the depth more or less follow the same distribution because it has only been trained with KITTI images. However the depth should not always be the same.

JIAJIAqi-1 commented 3 years ago

Thank you for your quick reply! I used the pretrained nets(dispnet_model_best.pth.tar) you posted on https://drive.google.com/drive/folders/1H1AFqSS8wr_YzwG2xWwAQHTfXN5Moxmx. And the input images are KITTI-rawdata-road like these 0000000000 0000000282

ClementPinard commented 3 years ago

Indeed there should be a better depth for these pictures. What size did you use ? Normally it should 416x128 . I guess you used the run_inference.py script ? Can you output the disparity instead of depth ? The resulting colored maps are usually more readable

JIAJIAqi-1 commented 3 years ago

Thank you very much! The reason of the problem is the incorrect size of pictures!

ClementPinard commented 3 years ago

great! :) Closing the issue then. Don't hesitate to reopen if you have further questions