Yifan Lu

I got it. I should download and run on the

@shubham-monarch replace the /colmap/dense/images with semantic maps before the stereo_fusion step

That's because I store the semantic map in the JPEG format. Actually, you need a png for the semantic image for lossless compression.

> @yifanlu0227 can you elaborate? > Are you suggesting replacing `(x,y,3)` to `(x,y,L,3)` for all the images? (`L` -> an integer label for each pixel in the image) > I...

> > That's because I store the semantic map in the JPEG format. Actually, you need a png for the semantic image for lossless compression. > > Could you comment...

> Let's say that I have `four` labels, namely `a`, `b`, `c` and `d`. We generate the semantic map for an image. The semantic map would have the same resolution...

Hi, that code snippet is deprecated, as well as `CoAlign/opencood/models/point_pillar_deform_transformer.py`.