-
Notifications
You must be signed in to change notification settings - Fork 16
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Low-shot training #5
Comments
Hi, We modify the code of DiffAug to train on the low-shot image generation task. |
Hi, would you mind sharing the source code of the low-shot generation experiments? It will help us a lot. |
Hello, |
Hi, I would add my vote to this discussion, it would be very helpful to have a look at the training script or the modified training.loss.py file. |
This experiment only shows in the supplementary and does not belong to the main paper. |
Hi, My modification to the DiffAugm was to add the lecam regularizatoin in the
I run the training for the same 300 kimg, I use
Thus, I am able to reproduce the original numbers from DiffAugm repository. However, the results after adding the lecam CR seem not to match to Table 4, this step is even harmful for 3/5 of the datasets. It would be indeed very helpful if we figure out where lies my misunderstanding. |
How to train the Lecam-gan on the low-shot image generation datasets,THX.
The text was updated successfully, but these errors were encountered: