Peize Sun

Hi~ We don't change detectron2(except adding CrowdHuman in dataset). You could install d2 and use sparsercnn directly.

Hi~ Yep, changing SCORE_THRESH_TEST/ INSTANCES_CONFIDENCE_THRESH in defaults.py won't change the result. This is because Sparse R-CNN only output 100/300 boxes in total for each image, so our implementation is putting...

Hi~ The proposal features contains information about its corresponding object. The proposal feature updates itself by interacting with RoI feature. We don't need feature information for each pixel position.

I guess Query is proposal features [100 x C], roi feature is Key [100 x (7 x 7 x C)] . Think about DETR, Query is object query [100 x...

Hi~ Suppose these many pictures are in the folder **demo_pictures**, You can run: `python demo/demo.py\ --config-file projects/SparseRCNN/configs/sparsercnn.res50.100pro.3x.yaml \ --input demo_pictures/* --output . --confidence-threshold 0.4 \ --opts MODEL.WEIGHTS path/to/model.pth`

Hi~ Our code is based on Detectron2, the way of using custom dataset is the same as detectron2, I guess you could refer to [Use Custom Dataset](https://detectron2.readthedocs.io/en/latest/tutorials/datasets.html).

Hi~ 150epoch is definitely better than 20epoch. However, you only get 1.6 mAP after 20epoch, it seems something wrong happens.

Hi~ We updateI the code several days ago. I guess now it is not "projects/SparseR-CNN", it is "projects/SparseRCNN"

I tried to reproduce your bug but I didn't reproduce it. I guess it maybe come from: (1) Your code is not the same as our updated code. (2) --input...

Hi~ In fact, our dynamic interaction is **not same** as multi-head attention. We **do not** recommend you understand dynamic interaction by the concepts of query, key and value.