Feng Wang

@kadirnar All method could be found in [`__init__.py`](https://github.com/Megvii-BaseDetection/YOLOX/blob/main/yolox/__init__.py), for YOLOX, no method is imported when `import yolox`; For yolov5 that you metioned, it could be found [here](https://github.com/fcakyon/yolov5-pip/blob/main/yolov5/__init__.py). As for cocoapi,...

Sorry that we don't have such a plan for detecting rotate object in official yolox. However, we encourage users to try it themselves and share code with the community.

You could patch the mask to the same shape and concat them. #Boxes in different image share the same logic like mask. You may refer these code in YOLOX and...

It's about definition of boxes. Ohter repo like detectron2 shares the same thing. You might refer to doc of detectron2 [here](https://detectron2.readthedocs.io/en/latest/notes/compatibility.html)

Usually too way to deal with negative coordinates: 1. clip to 0 2. keep the value unchanged Personally, I perfer the second way. If negative coordinates cause some bugs, the...

How about trying class agnostic nms? ( in demo_utils)

You may import `demo_postprocessing` from yolox.utils.demo_utils and just simply replace it with `postprocess`.

@ajwl27 sorry for not noticing your reply and in ```postprocess(prediction, num_classes, conf_thre=0.7, nms_thre=0.45, class_agnostic=False)```, setting `class_agnostic=True` could help. I mis-remember sth.

@rOtking We haven't tried YOLOX on overlapped object detection task. However, someone try yolox on crowd-human dataset and get a better result.

Could you plz provide a minor implementation ?