search

IntellicentAI-Lab / Re-GAN

[CVPR 2023] The official Pytorch implementation of Re-GAN
10 stars 0 forks source link

Pre-trained models or training configurations. #1

Open SerChirag opened 4 months ago

SerChirag commented 4 months ago

Hi, I was wondering if you have pertained models for this paper. I am looking for pre-trained models for few-shot datasets at 256×256 resolution (Obama, Panda, Grumpy-Cat, AnimalFace-Cat, AnimalFace-Dog).

I tried training your model from scratch a few times but I face mode collapse in most cases.

~ Chirag

SerChirag commented 4 months ago

Even the training configurations you used for each dataset would help.

JiiahaoXU commented 4 months ago

Hi Chirag,

Thanks for your interest in our work.

Unfortunately, I left HK PolyU so the checkpoints are probably deleted.

However, I ran the code contained in this repo last night, I didn't observe mode collapse (I tried the Panda dataset with the default setting in the repo except for the training batch size, I set it to 8).

BTW, if you are confused by model collapse, please consider using DiffAug to enhance the diversity of the model.

Thanks for your interest again.

Regards, Jiahao