Feng Li
Yes, DINO cannot do segmentation. Mask DINO is another paper to make DINO achieve good performance in segmentation tasks with three proposed methods. Please refer to our paper for more...
Hey, thanks for your interest in our paper. The paper is under review. We will release it after the blind review, which may need a few months.
Sorry I misunderstood your question. We do not use intance-level annotation of things for semantic segmentation. We do use bboxes in semantic segmentation (generated from all masks of the same...
Yes, your understanding is correct.
Hi, the hybrid matching is to match with classification loss, mask loss, and box loss (add the three losses together). Therefore, it is simple and we do not elaborate on...
Hey, we present these details in the appendix.
Sorry for the unclear description. In the appendix, we provide the loss function weights of different losses. The matching cost is the same as the loss weights.
Yes, you can just print the boxes of the selected boxes in query selection and visualize them. The quality of boxes after query selection is actually quite good.
Hey, I notice you are using Win11 os. We did not try compiling on Win11. Do your win11 have 4 or 8 gpus? Otherwise, you can not train this model....
Yes, it is enough for inference.