A replication of Denoising Diffusion Implicit Models paper with PyTorch and ViT.
To train a new model, you can modify the yaml file and:
python multi_gpu_trainer.py 20220822
Or you can download my pretrained weights for Oxford Flowers and run inference:
python ViT.py
The inference process is controled by 3 parameters (DiffusionVisionTransformer.sampler):
"device", usually torch.device('cuda') ;
"k", because the sampling become deterministic with DDIM, we can jump k steps for faster speed;
"N", how many samples you will get.
The result should looks like the welcoming images.
If you want my pretrained weights for miniImageNet, please leave you email address. Enjoy!