Architecture of Mask RCNN – A PyTorch Implementation

Re:Architecture of Mask RCNN – A PyTorch Implementation by edshiro on Saturday August 11, 2018 @10:25AM (#17739787)

Nice one! I don't remember all that much from reading the Mask-RCNN paper last year and have not seen many implementations so it's nice to be presented with this Pytorch implementation.

From what I recall about Faster R-CNN, the Regions Of Interest (ROI) are pre-determined via Selective Search, right? So I presume you would need to do the same thing with Mask-RCNN? I think this is the part I am the most confused with since I have never implemented Selective Search myself. Could you point me to introductory material on it?

Lastly, I can see the author of this work has read my blog post on understanding SSD MultiBox - glad it helped in some way :).

[ Parent | on HN ]

Re:Architecture of Mask RCNN – A PyTorch Implementation by HolyMakarony on Saturday August 11, 2018 @01:38PM (#17740667)

Another popular implementation of MRCNN in Kerala+Tensorflow is found here: https://github.com/matterport/Mask_RCNN.

Selective search is implemented in Fast-RCNN. Faster-RCNN improves upon that and uses a Region Proposal Me to propose RoI that may contain objects which speed up training and inference time.

[ Parent | on HN ]

Re:Architecture of Mask RCNN – A PyTorch Implementation by mliker on Saturday August 11, 2018 @11:20AM (#17739980)

RCNN uses selective search to generate the ROIs. What makes Faster RCNN faster is not having to spend time on selective search.

[ Parent | on HN ]

Re:Architecture of Mask RCNN – A PyTorch Implementation by wannabeOG on Saturday August 11, 2018 @12:21PM (#17740273)

It uses an RPN to generate the region proposals so it completely does away with the selective search which was the bottleneck for speed in fast RCNN

[ Parent | on HN ]