CLRNet
CLRNet copied to clipboard
Problem in generate_seg_tusimple.py
Hi, thanks for your nice work! When I read your codes and try to train in my own dataset, I find that the evaluation result is quite bad. My label is one-pixel. And when I read the generate_seg_tusimple.py, it seems the annotations have 30-pixel width. So there is the question: when you train, are the labels you used 30-pixel or 1-pixel? Thanks for your reply.