Tsinghua launches Event-STVG benchmark with 4,016 events for long-form video reasoning
Yizhe Zhang · hf · 2026-10-21
Researchers from Tsinghua University and Beijing University of Posts and Telecommunications introduce Event-STVG, a new task and large-scale benchmark that shifts spatio-temporal video grounding from locating instance-level targets to event-level reasoning.
- Requires long-form context, repeated-event disambiguation, role inference, and multi-target grounding
- Benchmark spans 4,016 annotated event occurrences across social and other domains
- State-of-the-art MLLMs struggle substantially, exposing gaps in long-form video reasoning
A notable new evaluation axis for multimodal understanding and video grounding research.
More from Research
- EvoMusArt 2027 opens submissions, due Nov 1, 2026 in Mainz, Germany — evomusart_conference · 2026-10-21
- AI denoising recovers superconductor spectra in 12s vs 5h, with error bars on the result — bravo_abad · 2026-10-21
- LLM2R recasts 3D part assembly as sequence modeling with a decoder-only LLM — NationalUniversityofSingapore · 2026-10-21
- Ant International's FlowState treats execution state as memory for long-horizon LLM agents — ant-intl · 2026-10-21
- Google's TEAL evolves reusable analytics tools for coding agents without ground-truth labels — google · 2026-10-21
- Tencent's ASAP evaluates agentic capabilities in pretrained base models before post-training — tencent · 2026-10-21