m-a-p/

IV-Bench

	IV‑Bench: A Benchmark for Image‑Grounded Video Perception and Reasoning in Multimodal LLMs




	
	Dataset Availability

Due to privacy policy, only a subset of the IV‑Bench dataset is publicly available. Specifically, we release 1,680 samples, including video IDs, image–text queries, and distractors.

Detailed usage instructions can be found on GitHub: IV‑Bench