this is my first time asking a question here, and I'm not an English native speaker. So please forgive my poor English.
I used matterport/MaskRCNN to train a model that can detect activities on the beach, like swim, surfing, and others. https://github.com/matterport/Mask_RCNN
But I got a really bad training result, it can be said that the model didn't learn anything.
After checking, I found that there is a problem when these codes generate the bbox from the mask.
The ground-truth visualization
The image includes two masks of swimming people. But after generating the bbox process, it only shows one bbox that includes two people. The bbox should be drawn respectivily one by one people.
I found the code named utils.extract_bboxes(mask), I understand it is the code that can generate bbox coordinate. But I have no idea why the problem appears when I use my dataset.
Is there any solution or code that can figure out this problem? or is there any explanation for why it will be a problem here?
The bbox generating code from utils.py (line 34, def extract_bboxes(mask)) https://github.com/matterport/Mask_RCNN/blob/master/mrcnn/utils.py
Thanks a lot.
