Summary of Generative Dataset Distillation Based on Diffusion Model, by Duo Su et al.
Generative Dataset Distillation Based on Diffusion Model
by Duo Su, Junjie Hou, Guang Li, Ren Togo, Rui Song, Takahiro Ogawa, Miki Haseyama
First submitted to arxiv on: 16 Aug 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary This paper presents a novel generative dataset distillation method based on Stable Diffusion, which can generate high-quality images at high speed. The authors focus on distillation methods based on the diffusion model, considering that the track requires generating a fixed number of images in 10 minutes using CIFAR-100 and Tiny-ImageNet datasets. They propose using the SDXL-Turbo model, which achieves an IPC (images per class) of 10 for Tiny-ImageNet and 20 for CIFAR-100, outperforming other diffusion models that can only generate IPC=1. The method uses class information as text prompts and post data augmentation for the SDXL-Turbo model to generate high-quality distilled datasets. Experimental results show the effectiveness of the proposed method, achieving third place in the generative track of the ECCV 2024 DD Challenge. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a new way to make computers generate pictures quickly and well. The authors use something called Stable Diffusion, which can make lots of good pictures fast. They want to make this happen for two types of pictures: CIFAR-100 and Tiny-ImageNet. They try different ways to do this and find one that works really well. This new way lets them make many more pictures than before (up to 20 times as many!). The authors tested their method and it did really well in a competition. |
Keywords
» Artificial intelligence » Data augmentation » Diffusion » Diffusion model » Distillation
