ReCon: Enhancing True Correspondence Discrimination through Relation Consistency for Robust Noisy Correspondence Learning

Benchmark Model Rank Results
cross-modal-retrieval-with-noisy-correspondence-on-cc152kReCon#1R-Sum: 380.5Image-to-text R@1: 43.1Image-to-text R@5: 68.7…
cross-modal-retrieval-with-noisy-correspondence-on-coco-noisyReCon#1R-Sum: 528.6Image-to-text R@1: 80.9Image-to-text R@5: 96.6…
cross-modal-retrieval-with-noisy-correspondence-on-flickr30k-noisyReCon#1R-Sum: 511.8Image-to-text R@1: 80.3Image-to-text R@5: 95.3…