Feng Li
Thanks for you quesiton. From my understanding, q_c should be in the first place, as the dimension matches.
Maybe the data has not been successfully loaded, like an error in loading annotations or doing augmentation. Can you give more details about your error?
As shown in ReadMe, `If your dataset files are not under this repo, you need to add export DETECTRON2_DATASETS=/path/to/your/data or use Symbolic Link ln -s to link the dataset into...
[ade_48.7log.txt](https://github.com/IDEA-Research/MaskDINO/files/12264000/ade_48.7log.txt) Hi, above is my log file for the `48.7` results for your reference. ADE20K is a small dataset, and the performance may not be so stable. I will also...
Hi, I think you only need to finetune the decoder part, it is lightweight and contains both detection and segmentaiton.
These weights are set with grid search. For example, class weight is set larger compared to DINO because mask losses will dilute the original class loss.
I am still trying to figure out this problem. If you wanna use `cuda:1` and run on this single GPU, you can add 'CUDA_VISIBLE_DEVICES=1' before your command, for example ```...
Maybe you guess if right, but I never used your way of specifying cuda devices. You can implement multiple maskdino models on multiple gpus with my command, which works well...
Yes, I believe there are logs in your saving directory.
Oh, it is a typo. Thanks for this comment, and you only need to load once.