Which .py file contains the main algorithm in your AAAI paper?

Question

Which .py file contains the main algorithm in your AAAI paper?

Newbeeer opened this issue 5 years ago · 7 comments

Newbeeer commented 5 years ago

I wonder about which file contains the LCCN training codes? Is it cifar10_train_varC.py?

Thanks

Answer 1 · 2019-04-13T05:31:00.000Z

Oh, yes. I will enrich the readme.

Answer 2 · 2019-04-13T05:43:37.000Z

Sorry to bother. I have some problems re-implemented your method in Pytorch.
How do you pre-train your ResNet32?

Answer 3 · 2019-04-13T05:45:13.000Z

Also, what's the data loaded for trans_warming ? train_init or eye(10)?

Answer 4 · 2019-04-13T10:17:26.000Z

Hi, I have updated the Readme. It can answer your question about the pre-training of ResNet32. For the trans_warming, we use the initialization mentioned in S-adaptation, i.e., the noise transition estimated by the M-step. It is pre-computed by our cifar10_train_T.py as a byproduct and saved in one pickle file.

Answer 5 · 2019-04-13T10:50:00.000Z

Hi, do you train the model for 120 epoch, and follow the 0.5-0.1-0.01 learning rate setting in pre-training?

Answer 6 · 2019-04-13T13:10:31.000Z

I use the model trained by CE to initialize the Forward, S-adaptation and LCCN models. This setting is indicated by the the argument init_dir as shown in the training command. So about your question, you can check the training procedure of CE.

Answer 7 · 2019-04-13T16:29:56.000Z

I see. Thank you