Mitigating Noisy Correspondence by Geometrical Structure Consistency Learning

Benchmark Model Rank Results
cross-modal-retrieval-with-noisy-1GSC-SGR#2R-Sum: 375.1Image-to-text R@1: 42.1Image-to-text R@5: 68.4
cross-modal-retrieval-with-noisy-2GSC-SGR#3R-Sum: 505.8Image-to-text R@1: 78.3Image-to-text R@5: 94.6
cross-modal-retrieval-with-noisy-3GSC-SGR#3R-Sum: 525.7Image-to-text R@1: 79.5Image-to-text R@5: 96.4