VisReflect: Latent Visual Reflection for Fine-Grained Perception in Long Visual Context

Benchmark Model Rank Results
visual-question-answering-on-v-benchVisReflect-7B-Image#10Accuracy: 84.9
video-question-answering-on-video-mme-with-subtitlesVisReflect-7B-Video#19Accuracy: 72.7
video-question-answering-on-mvbenchVisReflect-7B-Video#29Avg.: 70.4
visual-question-answering-vqa-on-blinkVisReflect-7B-Image#32Accuracy: 57.1
video-question-answering-on-mlvuVisReflect-7B-Video#33Accuracy: 70.6
zero-shot-video-question-answer-on-video-mme-w-o-subsVisReflect-7B-Video#33Accuracy (%): 67.4