Problems with training

[Luo-Ji-X]

Dear the authors:
I downloaded your code and trained it with 2 GTX1080Ti according to your readme document. But the training effect is much worse. Is there anything wrong with me? Attached is the training results.

           0        18        36        54  ...    144    162    180      mean

NM#5-6 0.937374 0.935354 0.948000 0.942424 ... 0.950 0.948 0.912 0.939596
BG#1-2 0.809091 0.829293 0.826263 0.835714 ... 0.816 0.840 0.787 0.811113
CL#1-2 0.780000 0.810000 0.819000 0.806122 ... 0.827 0.817 0.800 0.808069

[exitudio]

It's high variance due to the small data regime and noise from pose estimation. Especially, the "Coat" condition has the highest variance.
So we run 8 experiments for each architecture and report the best results.
We also provide the weights of the best result. You can find it here

[Tamako888]

It's high variance due to the small data regime and noise from pose estimation. Especially, the "Coat" condition has the highest variance. So we run 8 experiments for each architecture and report the best results. We also provide the weights of the best result. You can find it here

Did you use hyperparameter as same as the common.py setting default when running experiments, I really confused about the result I trained on a V-100. The mean accuracy is NM#5-6: 0.8499, BG#1-2: 0.7028, CL#1-2:0.6968

[exitudio]

Yes. We use the same default parameters as common.py. For BG and CL, the evaluation has very high variance but NM should not see much difference.
Can you run an evaluation on our pre-train model? Does it get a similar result as reported?