Save confidence scores of target domain as class-wise vectors about cbst

[yzou2]

Hi Hongwei, we downsample the score maps by 4 when saving them for later processing (ref: https://github.com/yzou2/CBST/blob/master/issegm/solve_ST.py#L864-L865). If your machine cannot handle it, you can increase the downsampling rate by 8 or 16. As far as I remember, it won't effect the results too much.

[Zhenghongwei0929]

Thank you very much for your response. It may be that my dataset is relatively larger than Cityscapse or GTA5, so it is difficult to directly store the probmap of all data. At present, I increased the downsampling rate and the problem was solved.