prlz77 / attend-and-rectify Goto Github PK

hi, i use your original code(it is wrn_28_10_ad=0_aw=0) to train cifar10, and get the best val accuracy is 95.99%, but when i use the attention version(the params are the same as ./log/wrn_28_10_ad=3_aw=4_err=3.45), the best val acc is 95.88%. it is even worse than the original code(without attention), could you explain it?

Thanks for the code you share! In the ECCV paper, it have shown that 11 convolution layer is used in attentional head but in the code it uses one 33 convolution layer to generate the attention heatmap. Could you please tell me why? Will it lead to better performance than 1*1 convolution?

Hi, thanks for sharing your work. But logs that record the process corresponding to results reported in paper are lost. Can you please share them again?