search

xuehy / Cross-View-Gait-Deep-CNNs

A Comprehensive Study on Cross-View Gait Based Human Idendification with Deep CNNs
45 stars 12 forks source link

test_set accuracy #4

Closed 853498747 closed 4 years ago

853498747 commented 5 years ago

Excuse me, I ran your code. The validation set works well, but why is the result of the test set much worse than the DeepCNNs paper?

xuehy commented 5 years ago

How many iterations did you run? Did you select the parameters when the validation accuracy is the highest?

853498747 commented 5 years ago

iteration = 300K and validation_set accuracy of 0.96. The test_set results are only a little better than the form you posted, and there is a big gap between the results in the DCNNS paper.

xuehy commented 5 years ago

My form is not from the final result. In your training procedure, is the validation accuracy still increasing? You may try to run more iterations. Someone has offered me feedbacks that the training takes a lot of iterations to converge.

853498747 commented 5 years ago

After iterations morethan 300K, the accuracy of the verification set can no longer be increased. I still have some confusion about them. I will ask you after I have a deep understanding of the code and the paper.Thank you for your patience.