Summary of Qvd: Post-training Quantization For Video Diffusion Models, by Shilong Tian et al.
QVD: Post-training Quantization for Video Diffusion Models
by Shilong Tian, Hong Chen, Chengtao Lv, Yu Liu, Jinyang Guo, Xianglong Liu, Shengxi Li, Hao Yang, Tao Xie
First submitted to arxiv on: 16 Jul 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 addresses the limitations of video diffusion models (VDMs) in generating coherent and realistic video content while maintaining computational efficiency and reduced memory consumption. The authors observe that temporal features exhibit pronounced skewness and inter-channel disparities, making it challenging to apply post-training quantization (PTQ) techniques. To address these issues, they introduce QVD, a PTQ strategy tailored for VDMs, comprising the HTDQ method for temporal features and SCRI for improving quantization level coverage across channels. Experimental results demonstrate the effectiveness of QVD in terms of diverse metrics, achieving near-lossless performance degradation on W8A8. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper helps make video diffusion models better by making them use less memory and be faster. The problem is that these models are really good at generating realistic videos but they’re also very big and take a long time to process. To solve this, the authors came up with a new way of reducing the size of these models while still keeping them working well. They tested their method on different models and datasets and showed that it works really well. |
Keywords
» Artificial intelligence » Diffusion » Quantization
