LLaVA-NeXT-Interleave: Tackling Multi-image, Video, and 3D in Large Multimodal Models

Benchmark Model Rank Results
zero-shot-video-question-answer-on-vnbenchLLaVA-NeXT-Video-7B#6Accuracy: 20.1
object-hallucination-evaluation-on-popeLLaVA-NeXT-Interleave 14B#7F1: 87.9
video-question-answering-on-next-qaLLaVA-NeXT-Interleave(14B)#15Accuracy: 79.1
video-question-answering-on-next-qaLLaVA-NeXT-Interleave(7B)#18Accuracy: 78.2
video-question-answering-on-next-qaLLaVA-NeXT-Interleave(DPO)#19Accuracy: 77.9
visual-question-answering-vqa-on-blinkLLaVA-NeXT-Interleave 14B#42Accuracy: 52.1
visual-question-answering-on-docvqa-testLLaVA-NeXT-Interleave 14B#73ANLS: 0.789
video-question-answering-on-mvbenchLLaVA-NeXT-Interleave 14B#76Avg.: 54.9
chart-question-answering-on-chartqaLLaVA-NeXT-Interleave 14B#881:1 Accuracy: 71.2