Views
No views yet
git clone https://github.com/mandt-lab/uqdm.git
cd uqdm
conda env create -f environment.yml
conda activate uqdm./data/imagenet64. Our implementation removes the duplicate test images as saved in ./data/imagenet64/removed.npy during loading.config.json and checkpoint.pt in a shared folder and load them for example via1from uqdm import load_checkpoint, load_data
2model = load_checkpoint('checkpoints/uqdm-tiny')
3train_iter, eval_iter = load_data('ImageNet64', model.config.data)1model.trainer(train_iter, eval_iter)
2model.evaluate(eval_iter)1image = next(iter(eval_iter))
2compressed = model.compress(image)
3reconstructions = model.decompress(compressed)1@article{yang2025universal,
2 title={Progressive Compression with Universally Quantized Diffusion Models},
3 author={Yibo Yang and Justus Will and Stephan Mandt},
4 journal = {International Conference on Learning Representations},
5 year={2025}
6}