Chinese CLIP: Contrastive Vision-Language Pretraining in Chinese

Open paper
Benchmark Model Rank Results
image-retrieval-on-coco-cnCN-CLIP (ViT-H/14)#1R@1: 81.5R@10: 99.1R@5: 96.9
image-retrieval-on-coco-cnCN-CLIP (ViT-L/14@336px)#2R@1: 80.1R@10: 99.2R@5: 96.7
image-retrieval-on-coco-cnCN-CLIP (ViT-L/14)#4R@1: 78.9R@10: 99.0R@5: 96.3
image-retrieval-on-coco-cnCN-CLIP (ViT-B/16)#5R@1: 77.0R@10: 99.0R@5: 97.1
image-retrieval-on-coco-cnCN-CLIP (RN50)#9R@1: 66.8R@10: 97.0R@5: 91.1
image-retrieval-on-flickr30k-cnCN-CLIP (ViT-L/14@336px)#4R@1: 84.4R@10: 98.7R@5: 97.1
image-retrieval-on-flickr30k-cnCN-CLIP (ViT-H/14)#5R@1: 83.8R@10: 98.6R@5: 96.9
image-retrieval-on-flickr30k-cnCN-CLIP (ViT-L/14)#6R@1: 82.7R@10: 98.6R@5: 96.7
image-retrieval-on-flickr30k-cnCN-CLIP (ViT-B/16)#7R@1: 79.1R@10: 97.4R@5: 94.8
image-retrieval-on-flickr30k-cnCN-CLIP (RN50)#11R@1: 66.7R@10: 94.1R@5: 89.4
image-retrieval-on-muge-retrievalCN-CLIP (ViT-H/14)#1Mean Recall: 83.6R@1: 68.9R@10: 93.1R@5: 88.7
image-retrieval-on-muge-retrievalCN-CLIP (ViT-L/14@336px)#2Mean Recall: 81.3R@1: 65.3R@10: 92.1R@5: 86.7
image-retrieval-on-muge-retrievalCN-CLIP (ViT-L/14)#3Mean Recall: 80.1R@1: 63.3R@10: 91.3R@5: 85.6
image-retrieval-on-muge-retrievalCN-CLIP (ViT-B/16)#5Mean Recall: 77.4R@1: 58.4R@10: 90.0R@5: 83.6
image-retrieval-on-muge-retrievalCN-CLIP (RN50)#7Mean Recall: 69.2R@1: 48.6R@10: 84.0R@5: 75.1

Rank counts only results with a code link.