search

tobyperrett / trx

Temporal-Relational CrossTransformers (CVPR 2021)
108 stars 23 forks source link

The number of iterations and the steps to drop learning rate? #6

Closed PHDJieFu closed 3 years ago

PHDJieFu commented 3 years ago

Thanks for your published work! When I reproduce your results on the Kinetics-100 dataset (with a total of 100k iterations), I find that the accuracy of the model will drop later. When will the performance of the model reach the best in the training process？ In addition, can you give me more details about the number of training iterations and the steps to drop the learning rate on both SSV2 and Kinetics-100 datasets. Thank you again for your help！

tobyperrett commented 3 years ago

Hi. You'd use the validation set to determine when to stop training ( https://github.com/tobyperrett/trx/issues/3 ). I experimented with dropping the learning rate, but it didn't make much of a difference (<1%), so wasn't worth the extra computation time/hyperparameter search.