On s3dis, those results are already behind SOTA

Question

On s3dis, those results are already behind SOTA

tchaton opened this issue 5 years ago · 1 comments

https://arxiv.org/pdf/1904.02113.pdf

Answer 1 · 2019-09-25T09:51:38.000Z

Hi, segments similar to super-pixels cast constraint of local labelling consistency. That's why SOTA or CRF-like networks sometimes outperform ours. However, you might notice that the 6-folds results of SOTA is actually on-par to ours. Compared to the computation cost of SOTA, the efficiency, scalability and generalisation of our network make it potentially more promising.