Lost in the Tail: Addressing Geographic Imbalance in Urban Visual Place Recognition

Benchmark Model Rank Results
visual-place-recognition-on-mapillary-valCliqueMining + LB loss–Recall@1: 94.7
visual-place-recognition-on-mapillary-valSALAD + LB loss–Recall@5: 96.5Recall@10: 97
visual-place-recognition-on-pittsburgh-30k-testCosPlace (ResNet101,128D) + LB loss–Recall@10: 97.9
visual-place-recognition-on-pittsburgh-30k-testImAge + LB loss–Recall@1: 94.3
visual-place-recognition-on-pittsburgh-30k-testSALAD + LB loss–Recall@5: 96.8
visual-place-recognition-on-sf-xl-test-v1DAPR-M (DINOv2, LB loss + CFD)–Recall@10: 92.5Recall@5: 92.4
visual-place-recognition-on-sf-xl-test-v1MegaLoc + LB loss–Recall@1: 95.4
visual-place-recognition-on-sf-xl-test-v2DAPR-M (DINOv2, LB loss + CFD)–Recall@5: 97Recall@10: 97.5
visual-place-recognition-on-sf-xl-test-v2ImAge + LB loss–Recall@1: 95.2