Fine-grained retrieval on Cars-196 (test)
92.2R@1Dense
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DensePre-training=SAE ImageNet-1K, Evaluation Protocol=zero-shot2026.03 | 92.2 | 98.8 | 99.6 | |
| FAISS-HNSWPre-training=SAE ImageNet-1K, Evaluation Protocol=zero-shot2026.03 | 92.2 | 98.8 | 99.6 | |
| Two-stage BM25-V pipelinePre-training=SAE ImageNet-1K, Evaluation Protocol=zero-shot, Candidate pool size (K)=2002026.03 | 91.8 | 98.5 | 99.3 | |
| Two-stage BM25-V pipelinePre-training=SAE ImageNet-1K, Evaluation Protocol=zero-shot, Candidate pool size (K)=1002026.03 | 91.4 | 98.3 | 99 | |
| FAISS-IVF+PQPre-training=SAE ImageNet-1K, Evaluation Protocol=zero-shot2026.03 | 89.7 | 97.8 | 98.7 | |
| BM25-VPre-training=SAE ImageNet-1K, Evaluation Protocol=zero-shot2026.03 | 71.5 | 90 | 94.6 |