CVPR-NAS 2021: Unseen Data Track Forum

Go back to competition Back to thread list Post in this thread

> except ingestion, training question

after searching the all
since training searched model with nas takes much more time than searching (ingestion itself)
results = torch_evaluator(model, data, metadata, n_epochs=64, full_train=full_train)
can we modify n_epochs to lower number such as 32 in evaluation submission or test submission to run nas within 12 hours ? (without changing optimizer or other things just epochs)

Posted by: ChoiSangbum @ May 1, 2021, 1:01 p.m.


No, sorry. We want to keep the training for the develpment phase constant throughout; all submisssions thus far have been evaluated over 64 epochs. For the next phase, we can extend the 12 hour time if people are running into issues evaluating their algorithms over the new datasets.


Posted by: robgeada @ May 1, 2021, 1:19 p.m.
Post in this thread