modifying MODEL.MASK_FORMER.NUM_OBJECT_QUERIES can't load pretrained model when training about mask2former HOT 3 CLOSED

The pre-trained model won't work if MODEL.MASK_FORMER.NUM_OBJECT_QUERIES is modified. You will need to train the model again if you want to use different MODEL.MASK_FORMER.NUM_OBJECT_QUERIES.

from mask2former.

@bowenc0221 Thanks for your answer.

But I got this error.

RuntimeError: Error(s) in loading state_dict for MaskFormer:
        size mismatch for sem_seg_head.predictor.query_feat.weight: copying a param with shape torch.Size([100, 256]) from checkpoint, the shape in current model is torch.Size([30, 256]).

I use pre-trained model weights to train my custom dataset, and I modified MODEL.MASK_FORMER.NUM_OBJECT_QUERIES = 30 , but it dosen't work. So you mean that I can't use pre-trained model to fine tune on my custom dataset?

from mask2former.

As the error suggests, the pre-trained model you use has 100 queries instead of 30.

from mask2former.