Denoising Diffusion Probabilistic Models, Jonathan Ho, Ajay Jain, and Pieter Abbeel, 2020Advances in Neural Information Processing Systems (NeurIPS)DOI: 10.48550/arXiv.2006.11239 - Foundational paper introducing the Denoising Diffusion Probabilistic Models, outlining the iterative refinement process that forms the basis of diffusion model inference.
Denoising Diffusion Implicit Models, Jiaming Song, Chenlin Meng, and Stefano Ermon, 2020International Conference on Learning Representations (ICLR)DOI: 10.48550/arXiv.2010.02502 - Introduces Denoising Diffusion Implicit Models (DDIMs), offering a faster and more efficient sampling strategy compared to DDPMs, directly addressing the bottleneck of numerous sampling steps.
U-Net: Convolutional Networks for Biomedical Image Segmentation, Olaf Ronneberger, Philipp Fischer, and Thomas Brox, 2015Medical Image Computing and Computer-Assisted Intervention (MICCAI), Vol. 9351 (Springer, Cham)DOI: 10.1007/978-3-319-24574-4_28 - Presents the U-Net architecture, which is the core neural network repeatedly evaluated in diffusion models and a primary source of computational cost during inference.
High-Resolution Image Synthesis with Latent Diffusion Models, Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, Björn Ommer, 2022Conference on Computer Vision and Pattern Recognition (CVPR)DOI: 10.48550/arXiv.2112.10752 - Introduces Latent Diffusion Models (LDMs) like Stable Diffusion, which operate in a compressed latent space to enable high-resolution image generation, highlighting the computational scale of modern diffusion models.
Progressive Distillation for Fast Sampling of Diffusion Models, Tim Salimans and Jonathan Ho, 2022International Conference on Learning Representations (ICLR)DOI: 10.48550/arXiv.2202.00512 - Proposes a distillation method to enable diffusion models to produce high-quality samples with significantly fewer sampling steps, directly addressing the bottleneck of repetitive U-Net evaluations.