Closed wangSir0202 closed 4 weeks ago
@wangSir0202 Firstly, the method is only designed for CZSL task, not for detection or segmentation. However, I think you can train a recognition model on a pre-processed COCO dataset, which contains images with state-object compositional labels, if it is neccessary.
@wangSir0202 I believe the problem has been resolved, so I am closing this issue. Feel free to reopen it if needed.
Dear author, may I have a question that how can I have a train on COCO dataset?