question about the performance on my own data

[jucic]

Thanks for your nice job.
I tried the pretrained model on my own data, however the disparity result looks not good because of the blurry depth edge and
incorrect depth value, which is not the same with the result in following link:
https://varstuff.s3-eu-west-1.amazonaws.com/highresstereo/viewer.html
I've set maxdisp to 80 according to the disparity threshold of my own data. Is it possible that anything else went wrong?

[gengshan-y]

If the pre-trained model works on the demo image (crusade), then I'm guessing the problem might be related to image resolution.

If would help to know the image resolution being used, and what kind of incorrect depth patterns are observed.

Would it help to upsample the image by passing --testres res, where res could be 2 or even 4?