I am trying to use detcetron2 panoptic_FPN and panoptic_deeplab models for optimization on custom dataset.
You might already now that coco has around 133 classes (both thing and stuff). And my custom dataset contains 7 classes ( both thing and stuff), out of which 5 classes are already present in coco and 2 classes are new.
I tried to train panoptic_FPN on custom dataset with using coco-pretrained weights and my custom classes which are mapped to the categories id's of coco-classes. but when i tried to test the newly obtained weights on test data, i could only see the predictions of my custom classes. The predicitons of the coco-classes are available. what is the mistake i am doing in trainig ?.
yes, I did change the following values in config file: MODEL.ROI_HEADS.NUM_CLASSES = 135 (133 coco + 2 new custom classes) MODEL.SEM_SEG_HEAD.NUM_CLASSES = 136 (N+1)
I would like to know whether, i need to retrain the model combinig both the coco and custom dataset. to make use of both the model classes
