Recall@k Surrogate Loss with Large Batches and Similarity Mixup

Benchmark Model Rank Results
image-retrieval-on-inaturalistRecall@k Surrogate loss (ViT-B/16)#3R@1: 83.0R@16: 95.9R@32: 97.2R@5: 92.1
image-retrieval-on-inaturalistRecall@k Surrogate loss (ResNet-50)#5R@1: 71.8R@16: 91.9R@32: 94.3R@5: 84.7
metric-learning-on-cars196Recall@k Surrogate loss (ViT-B/16)#9R@1: 89.5
metric-learning-on-cars196Recall@k Surrogate loss (ResNet-50)#15R@1: 88.3
metric-learning-on-stanford-online-products-1Recall@k Surrogate Loss (ViT-B/16)#3R@1: 88.0
metric-learning-on-stanford-online-products-1Recall@k Surrogate Loss (ViT-B/32)#7R@1: 85.1
metric-learning-on-stanford-online-products-1Recall@k Surrogate Loss (ResNet-50)#10R@1: 82.7
vehicle-re-identification-on-vehicleid-largeRecall@k Surrogate loss (ViT-B/16)#1Rank-1: 94.7Rank-5: 97.1
vehicle-re-identification-on-vehicleid-largeRecall@k Surrogate loss (ResNet-50)#2Rank-1: 93.8Rank-5: 96.6
vehicle-re-identification-on-vehicleid-mediumRecall@k Surrogate loss (ViT-B/16)#1Rank-1: 95.2Rank-5: 97.2
vehicle-re-identification-on-vehicleid-mediumRecall@k Surrogate loss (ResNet-50)#2Rank-1: 94.6Rank-5: 96.9
vehicle-re-identification-on-vehicleid-smallRecall@k Surrogate loss (ViT-B/16)#1Rank-1: 96.2Rank-5: 98.0
vehicle-re-identification-on-vehicleid-smallRecall@k Surrogate loss (ResNet-50)#2Rank-1: 95.7Rank-5: 97.9