Preliminary understanding of segmentation research
Last few months, I was doing a lot of work based on GOTURN, which is a tracking algorithm by deep learning and trained end to end. I tried to reproduce this work from scratch, but finally my model has performance gap with origin one.(0.177 vs 0.205) This reproduce work has already taken a very long time for my unenlightened knowledge on deep learning and visual tracking. In this process, I still have some gains, about how to implement a project or how to check your code is right.
Util April, I switched into another baseline, OSMN, a segmentation project for DAVIS benchmark. This work is really interesting. So I begin to record my gains and pains during this process.