Continuous-Time Distribution Matching for Few-Step Diffusion Distillation
About this episode
This episode discusses Continuous-Time Distribution Matching for Few-Step Diffusion Distillation, a novel approach to enhance diffusion models.
🤗 Upvotes: 24 | cs.CV, cs.AI Authors: Tao Liu, Hao Yan, Mengting Chen, Taihang Hu, Zhengrong Yue, Zihao Pan, Jinsong Lan, Xiaoyong Zhu, Ming-Ming Cheng, Bo Zheng, Yaxing Wang Title: Continuous-Time Distribution Matching for Few-Step Diffusion Distillation Arxiv: http://arxiv.org/abs/2605.06376v1 Abstract: Step distillation has become a leading technique for accelerating diffusion models, among which Distribution Matching Distillation (DMD) and Consistency Distillation are two representative paradigms. While consistency methods enforce self-consistency along the full PF-ODE trajectory to steer it toward the clean data manifold, vanilla DMD relies on sparse supervision at a few predefined discrete timesteps. This restricted discrete-time formulation and mode-seeking nature of the reverse KL divergence tends to exhibit visual artifacts and over-smoothed outputs, often necessitating complex auxiliary modules -- such as GANs or reward models -- to restore visual fidelity. In this work, we introduce Continuous-Time Distribution Matching (CDM), migrating the DMD framework from discrete anchoring to continuous optimization for the first time. CDM achieves this through two…
More episodes of Daily Paper Cast
- MiA-Signature: Approximating Global Activation for Long-Context Understanding
- When to Trust Imagination: Adaptive Action Execution for World Action Models
- Stream-R1: Reliability-Perplexity Aware Reward Distillation for Streaming Video Generation
- Stream-T1: Test-Time Scaling for Streaming Video Generation
- RLDX-1 Technical Report
- OpenSearch-VL: An Open Recipe for Frontier Multimodal Search Agents
Explore listener stats, chart rankings, contacts and more on the Daily Paper Cast podcast page.