LAION-BVD: A 10-Million-Hour Open Video Dataset for Multimodal Pre-training

Benchmark Model Rank Results
zero-shot-action-recognition-on-hmdb51ViCLIP L-14 (BVD-V-50M, zero-shot)–Top-1 Accuracy: 61.4
zero-shot-action-recognition-on-ucf101ViCLIP L-14 (BVD-V-55M, zero-shot)–Top-1 Accuracy: 80.5
zero-shot-transfer-image-classification-on-imagenet-rCLIP ViT-B/16 (LAION-BVD, zero-shot)–Accuracy: 40
zero-shot-video-retrieval-on-msr-vttViCLIP L-14 (BVD-V-50M, WiSE-FT, zero-shot)–text-to-video R@1: 44.6
zero-shot-video-retrieval-on-msr-vtt-video-to-textViCLIP L-14 (BVD-V-10M, WiSE-FT, zero-shot)–R@1: 45.3