Efficient image generation with continuous label control and fewer steps
Efficient Covariance-Aware Vicinal Latent Diffusion for Continuous Conditional Image Generation
Computer Vision and Pattern Recognition
Summary
Generating images based on numbers that change smoothly, like age or angle, is hard when the data used to train the computer is unbalanced or rare for some values. The authors introduce a new method called CAViLD to build a smarter way for the computer to understand and generate images in a smaller, meaningful space. Their approach also speeds up generating images by learning how to guide the process without extra parts and needing fewer steps. Tests show that CAViLD creates better, more consistent images while working faster than some older methods.
What this means in practice
- •For graphic designers: Produce images that smoothly change according to continuous parameters like age or angle with improved speed and quality.
- •For autonomous vehicle software teams: Generate diverse training images conditioned on steering angles efficiently, improving simulation and model training processes.
Authors
Xin Ding, Shuo Huang, Zhenglong Ding, Sen Zhang, Nenglun Chen, Peibei Cao, Yongwei Wang, Fei Wu
Abstract
Continuous Conditional Generative Modeling (CCGM) aims to synthesize high-dimensional data conditioned on continuously valued scalar labels. However, existing continuous diffusion models face severe data sparsity under imbalanced and long-tailed label distributions, which can lead to visual artifacts and condition inconsistency. Moreover, inference-time classifier-free guidance and dense pixel-space integration introduce substantial sampling overhead, limiting practical deployment. To address these challenges, we propose CAViLD, an efficient latent diffusion framework for continuous conditional generation. CAViLD first constructs a purified condition-aware latent space through density-adaptive vicinal learning, auxiliary semantic regularization, and NIQE-guided negative data augmentation. It then internalizes the conditional guidance direction through an EMA teacher, enabling single-branch conditional inference, and introduces covariance-aware progressive distillation for label-dependent probability-flow ODEs, supporting sampling down to four NFEs. Experiments on RC-49, UTKFace, and Steering Angle demonstrate a favorable trade-off among generation fidelity, continuous condition consistency, and sampling efficiency.