Open swords123 opened 4 years ago
@swords123 Unlike NYU Depth v2 dataset, each scene of KITTI dataset has focal length different to each other. So, for testing with scenes from KITTI dataset, we finally scale depth estimations with focal value divided by average focal length computed across the training set.
I have found that you have this line in your code, can you explain why it need to do it, thank you!
final_depth = final_depth * focal.view(-1, 1, 1, 1).float() / 715.0873