Experiments results

[Yuranus]

Thank you for sharing the code. I also re-implement the algorithm but can't obtain approximate results like in the paper. I wonder if there are some tricks in training but were not mentioned in the original paper.

[luanyunteng]

In fact, I faced the same problem. In my opinion, the paper's some details are not clear, such as the side-branch bottleneck architectures.