Taming Hallucinations: Boosting MLLMs' Video Understanding via Counterfactual Video Generation

Open paper

Benchmark	Model	Rank	Results
video-question-answering-on-tempcompass-overall	DNA-Train-7B	#7	Accuracy: 73.5
video-question-answering-on-tomato	DNA-Train-7B	#10	Accuracy: 32.6
video-question-answering-on-mvbench	DNA-Train-7B	#54	Avg.: 63.8