Tensorflow code is "L-GM-Loss" or "L-Center-Loss"

[Hc-Xie]

Hello, I am very interested in this research project and I believe what you demonstrated in the paper is novel. However, I am not sure is this the same in your original caffe's code or not but in your tensorflow code, line 451-454 of resnet_model.py, what you are doing is "Margined Center Loss" instead of "Margined GM loss". There is no covariance involved in your calculating code (or your covariance matrix is identity matrix) which I believe is completely different with what you showed in the paper.

[nothingeasy]

I have the same question about it. Could any one explain the implementation ?