Running priordata test doesn't work on users own data but works with sample data.

[DrMaxP]

I have the sample working on Windows, once the output filepath format has been adjusted to be compatible (to change the : in the time to another character) - e.g.

priorda test --image_path assets/sample-1/rgb.jpg --prior_path assets/sample-1/gt_depth.png --pattern downscale_32 --visualize 1

I'm now trying to run the same on my own data with depth priors. I have a colour jpg and depth jpg that are the same dimension, but considerably higher resolution than the sample (at 4096x3072). I see a runtime error:RuntimeError: shape mismatch: value tensor of shape [37748736] cannot be broadcast to indexing result of shape [9216]

Are there any restrictions on the input image formats or resolution?

Thanks,

Max

[MaxPalmer-UH]

OK, I have got my own data working. There were a couple of things to note:

1. My images are likely RGBA, I had to change the image loading code to force a conversion to RGB.
2. For some reason, if the rgb and depth prior image are not called rgb.png and gt_depth.png the code fails (on output). It seemed I couldn't use my own names and pass those into the code.

Upshot is, this is working now on Windows 11 - but it required some very minor changes.