sahaj432
commented
4 years ago Hi Warday, I am at an intermediate level in deep learning. From my point of view, the answer is "you need to have another network and you need to retrain". Because, if you don't want the segmentation part in the Mask RCNN algorithm then it just becomes a faster RCNN algorithm. The answer to computational speed the research of Mask RCNN has mentioned that by adding segmentation part (FCN) in Faster RCNN there is little raise in computational time for detection. So, from my point of view, you should go with a faster RCNN algorithm as you don't need masking.
Hopefully, this would answer your question.
Hi, I have been working with maskrcnn to detect and segmentation leaves on trees. With several test we decide to use only de detection (box) and discard the segmentation. To increase performance, I would like to know if the segmentation process can be ignore to speed up the detection or I will have to retrain another network. The problem is that the training and validate database have been prepared to train this network and I doubt that I can use it to train an object detection network. Thenx!