Read the original at HF Daily Papers
Researchers introduce OmniCompass, a benchmark for evaluating multimodal models on video grounding and understanding. It includes 820 vision-language videos and 240 audio-visual videos.
Carried by: HF Daily Papers. First seen: .