Lingsi Zhu

Resolving visually indistinguishable human micro-actions requires more than raw foundation model scale: hierarchical coarse-to-fine soft fusion paired with targeted candidate reranking pushes long-tailed micro-action recognition to a benchmark-winning 79.99% F1-mean.

ACE-Brain-0.5 unifies spatial reasoning and action generation in embodied AI, achieving remarkable performance improvements across multiple benchmarks.

By using text as an anchor, this model achieves state-of-the-art emotional mimicry intensity estimation, even when visual and acoustic data are noisy or missing.

A dual-branch Transformer with safe cross-attention overcomes missing visual cues in emotion recognition by dynamically relying on audio, achieving state-of-the-art results on Aff-Wild2.