VideoJev: System One Model for Video Understanding and Interaction

Benchmark Model Rank Results
temporal-sentence-grounding-on-activitynet-timelensVideoJev-35B-A3B#1mIoU: 65.5R@1 IoU=0.3: 81.0R@1 IoU=0.5: 72.5R@1 IoU=0.7: 54.4
temporal-sentence-grounding-on-activitynet-timelensVideoJev-4B#3mIoU: 62.8R@1 IoU=0.3: 78.8R@1 IoU=0.5: 70.5R@1 IoU=0.7: 51.4
temporal-sentence-grounding-on-charades-timelensVideoJev-35B-A3B#1mIoU: 63.7R1@0.5: 74.8R1@0.7: 44.5R1@0.3: 86.1
temporal-sentence-grounding-on-charades-timelensVideoJev-4B#3mIoU: 60.1R1@0.5: 69.4R1@0.7: 40.3R1@0.3: 82.8
temporal-sentence-grounding-on-qvhighlights-timelens-validationVideoJev-35B-A3B#1mIoU: 74.0R@1 IoU=0.3: 88.0R@1 IoU=0.5: 79.9R@1 IoU=0.7: 65.5
temporal-sentence-grounding-on-qvhighlights-timelens-validationVideoJev-4B#2mIoU: 73.2R@1 IoU=0.3: 87.3R@1 IoU=0.5: 79.4R@1 IoU=0.7: 64.8
zero-shot-video-question-answer-on-video-mme-w-o-subsVideoJev-35B-A3B#5Accuracy (%): 81.8
zero-shot-video-question-answer-on-video-mme-w-o-subsVideoJev-4B#11Accuracy (%): 75.0
video-question-answering-on-mlvuVideoJev-35B-A3B#6Accuracy: 85.9
video-question-answering-on-mlvuVideoJev-4B#11Accuracy: 80.6
video-question-answering-on-lvbenchVideoJev-35B-A3B#9Accuracy: 73.4
video-question-answering-on-lvbenchVideoJev-4B#14Accuracy: 67.9