Rectified Flow Enhances Emotion Estimation
Salah Eddine Bekhouche · hf · 2026-07-16
AffectFlow-DINO introduces a multi-task emotion estimation system for the ABAW challenge, adding a conditional rectified flow head to traditional deterministic architectures to capture the inherent ambiguity of facial behavior annotations in real-world scenarios.
Core Method
- Instead of predicting a single outcome, it learns a conditional generative distribution, enabling one-to-many uncertainty predictions via Monte Carlo sampling.
- Simultaneously tackles three tasks: continuous valence-arousal regression, 8-class expression classification, and 12 Action Units detection.
- Utilizes frozen DINOv3 ViT-S/16 as the backbone.
Results
- Extensive ablations show that rectified-flow decoding consistently improves deterministic predictions, especially valence-arousal: CCC-V increased by 0.058.
- Posterior threshold calibration significantly fixes long-tail rare category performance without retraining, boosting Fear from 3.8% to 33.1%.
- Combining backbone fine-tuning and flow retraining achieves a final PMTL = 1.177, substantially outperforming the official baseline of 0.45.
More from Research
- Stanford Team Introduces Gigatoken, the World's Fastest Tokenizer — StanfordAILab · 2026-07-22
- Tabul AI launches Metal TreeSHAP to speed up Shapley values on Apple silicon — Scobleizer · 2026-07-22
- Reddit points to OpenAI’s ChatGPT Ads page — EcstaticAsparagus509 · 2026-07-22
- Open-source runtime lets each repo define its own AI code reviewer — ibabufrik · 2026-07-22
- DeepSWE: A New Benchmark for Evaluating AI Coding Agents on Real GitHub Issues — pmz · 2026-07-22
- A Rust space-economy sim runs hundreds of autonomous ships, built with Claude — kalcode · 2026-07-22