EVA-Cap: Fudan and OPPO align video captioning with atomic audiovisual events via GRPO
Fudan-University · hf · 2026-10-16
Fudan University's Institute of Trustworthy Embodied AI, with OPPO's AI Center, proposed EVA-Cap, addressing how Omni Language Models treat audiovisual captioning as unstructured text generation and miss videos' semantic structure.
- Decomposes captions into atomic audiovisual events, turning unstructured text supervision into event-centric alignment signals.
- Trains via supervised fine-tuning and then Group Relative Policy Optimization (GRPO) against three event-centric metrics on a self-curated, quality-controlled dataset, EVA-Data, improving fine-grained audiovisual alignment.
More from Multimodal
- OpenAI Astra tutorial shows PolyHaven-powered 3D scenes as GPT-6 docs page surfaces — jason_mayes · 2026-10-16
- Meituan LongCat's WorldCast benchmark: GPT-Image-2.5 tops at 72.8, open-source lags — meituan-longcat · 2026-10-16
- Three resources to get up to speed with FLUX 3 — ariG23498 · 2026-10-16
- Watch: Claude Opus 5.5 builds a Tesla Roadster from a single prompt — TAbrodi · 2026-10-16
- Midjourney v8.2 turns out striking indigo ink-wash art with sref 1848359068 — azed_ai · 2026-10-16
- "AI television from another dimension": robot grooms a chihuahua on air — BotLordBotflix · 2026-10-16