Training for segmentation

[hugozanini]

Hi, I'm trying to train pix2pix for segmentation. I'm using SUN-RGBD dataset (segmentation in scenes indoor), but my results are very bad. I believe that the problem may be in the consistency of the data. Sometimes things that we consider to have the same class (as wall) have different pixel values in different images. Can this be a problem?

[wxh001qq]

sorry, may i ask you how to run the evaluate.py? i found there request labels,and there are many fold that could not found in the cityscape dataset that the download.sh provide