DiffoR: A Unified Continuous Generative Framework for Universal Ordinal Regression

In three linesDiffOR introduces a novel paradigm for Ordinal Regression as Continuous Generative task. The framework leverages diffusion models to recover continuous ordinal values via iterative denoising, with a Dual-Decoupling Strategy (Multi-scale Increment Aggregation and Dynamic Denoising Perception) to preserve ordinal topology. Validated on 12 benchmarks across four domains.

Summary generated by Claude — human-verified