ViDiT-Q: Efficient and Accurate Quantization of Diffusion Transformers for Image and Video Generation.
Tianchen ZhaoTongcheng FangEnshu LiuRui WanWidyadewi SoedarmadjiShiyao LiZinan LinGuohao DaiShengen YanHuazhong YangXuefei NingYu WangPublished in: CoRR (2024)
Keyphrases
- input image
- image data
- image retrieval
- image features
- image classification
- quantization error
- edge detection
- image content
- image representation
- computationally efficient
- single image
- multimedia
- image frames
- visual data
- segmentation method
- image segmentation
- video data
- image regions
- anisotropic diffusion
- diffusion process
- video sequences
- multiscale
- video files
- image compression
- video frames
- multimedia data
- low level
- transform coefficients
- pixel domain
- quantization noise
- quantization step