Hi, first, I must recognize that it is an excellent paper. I reached this because I've been stuck trying to perform the inference and draw the unknown objects for at least two weeks.
By reviewing the repository, when you run the apply_net.py, two JSON files are created, one for the known objects and the other that includes the unknown objects. Based on that, I've been trying to run the inference of a single image and draw the bounding boxes of the known and unknown objects. The first ones are easy to get, but I need help with the unknown objects. I already explore the methods 'OOD_detection_postprocessing,' 'probabilistic_detector_postprocess,' 'mergeIDOOD,' etc. I tried to use the scores and complete scores to get the unknown, but if some unknown objects are detected, most unknown objects overlap with the known ones. What do I need to include to identify the unknown objects?
Hi, first, I must recognize that it is an excellent paper. I reached this because I've been stuck trying to perform the inference and draw the unknown objects for at least two weeks. By reviewing the repository, when you run the apply_net.py, two JSON files are created, one for the known objects and the other that includes the unknown objects. Based on that, I've been trying to run the inference of a single image and draw the bounding boxes of the known and unknown objects. The first ones are easy to get, but I need help with the unknown objects. I already explore the methods 'OOD_detection_postprocessing,' 'probabilistic_detector_postprocess,' 'mergeIDOOD,' etc. I tried to use the scores and complete scores to get the unknown, but if some unknown objects are detected, most unknown objects overlap with the known ones. What do I need to include to identify the unknown objects?