custom dataset without 'stuffthingmaps'
cjfcsjt opened this issue · comments
Hi!
I want to train this model on instance segmentation with mycustom dataset. I convert my dataset to the coco2017 format, but I do not have the 'stuffthimgmaps', I wonder if this is necesesary? If true, could you please give me some advice to create 'stuffthingmaps' using my custom dataset?
I revised the config file with reference to './configs/htc/htc_without_semantic_r50_fpn_1x.py', and it do not need the 'stuffthingmaps' now.