3-px error is always 0.000 on own dataset

[hubery4]

Hi @JiaRenChang ,

When I use my dataset to train the network, the 3-px error is always 0.000 and the output disparity is black using the model.
Do you have any ideas to handle this problem?

Thank you very much.

[hubery4]

Hi @JiaRenChang ,

When I use my dataset to train the network, the 3-px error is always 0.000 and the output disparity is black using the model.
Do you have any ideas to handle this problem?

Thank you very much.

@JiaRenChang @mileyan @H1Gdev
Also, when I use my own data to train main.py, the loss value will immediately converge and remain unchanged.

Then, when I test my own image use the pretrained model, the disparity output will be black.

Do you know how to fix it?

Thank you very much.

[JiaRenChang]

@hubery4
You can print your output tensor to see the values.
If the values seem correctly, the black disparity may be caused by saving .png.
You can change the save png tools, such as cv2.
Jia-Ren

[zzr0127]

Hello, I also have this problem, how to solve it？