one question about the predict value about omnimvs_pytorch

[matsuren]

Thank you for your comment.

Could you check if gt_idepth has the collect value? Because gt_invd_idx is calculated by the following code
gt_invd_idx = converter.invdepth_to_index(gt_idepth)

Thank you,

[matsuren]

Also, please consider using the official implementation available here.

[ouhuajing]

Thank you very much for sharing.
I encountered a problem that the resulting predicted disparity maps were completely black, while training the data without pre-trained weight file. By printing the data, I found that all the data in tensor "gt_invd_idx" from training dataset were 0, and the data in tensor "pred" which is the output of network model was 0~1 . Can I ask what is wrong? Thank you!

I encountered the same problem and the gt_idepth is so small less than 0.01. Do you have a solution ?

[matsuren]

Could you visualize gt_idepth using something like plt.imshow to make sure that you load dataset correctly? If all values are small, I guess you failed to load dataset.

[ouhuajing]

Could you visualize gt_idepth using something like plt.imshow to make sure that you load dataset correctly? If all values are small, I guess you failed to load dataset.

7/5000
Thank you for your reply，I revise the invdepth load way in the function load_invdepth(),
invdepth=invd_value/ 255.0 * 1.818 + np.finfo(np.float32).eps
now I can train the network.

[Lsm67]

Thanks a million to both of you for sharing.
Referring to your suggestions, I have finished a training model for these days. After modifying the function load_invdepth(),invdepth_to_index(),index_to_invdepth, I trained a model on sunny dataset for 200 epochs, 900 training pictures, although the evaluation "a1" is 54.8 .
The final disparity map is not as good as the one which the author shows. Maybe because the training data set is not enough.
thank you!!!

[ouhuajing]

Thanks a million to both of you for sharing.
Referring to your suggestions, I have finished a training model for these days. After modifying the function load_invdepth(),invdepth_to_index(),index_to_invdepth, I trained a model on sunny dataset for 200 epochs, 900 training pictures, although the evaluation "a1" is 54.8 .
The final disparity map is not as good as the one which the author shows. Maybe because the training data set is not enough.
thank you!!!

Thank you very much. Could you please give the modified function code?

[matsuren]

Hi @Lsm67,

Thank you for checking. As you mentioned, the number of disparity is smaller than the original paper because of #1.

Thanks,

[matsuren]

I think it's ok to close now.