VidUEU-Agent: A Data-Curation Agent for Multimodal Understanding, Editing, and Unified Tasks
The researchers built 100,000 training samples from raw video, with an agent that adjusts the data mix using downstream training results.
VidUEU-Agent filters low-quality shots, extracts keyframes and multimodal signals, then builds metadata to generate instruction samples. The resulting VidUEU-100K dataset covers multimodal understanding, editing and unified tasks across multiple domains. The authors report consistent benchmark gains after fine-tuning baseline models on it, though the abstract gives no numerical results.
HF Daily Papers' note
VidUEU-Agent filters low-quality shots, extracts keyframes and multimodal signals, then builds metadata to generate instruction samples. The resulting VidUEU-100K dataset covers multimodal understanding, editing and unified tasks across multiple domains. The authors report consistent benchmark gains after fine-tuning baseline models on it, though the abstract gives no numerical results.
HF Daily Papers' note
score 4