Image-to-image retrieval on ImageNet
57.6mAPOriginal
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OriginalBackbone=DINOv3, ViT=L/16, Similarity=⟨I, I⟩2026.03 | 57.6 | — | |
| PCA←Backbone=SigLIP2, ViT=B/16, Similarity=⟨I←, I←⟩2026.03 | 46.5 | — | |
| PCA←Backbone=SigLIP, ViT=B/16, Similarity=⟨I←, I←⟩2026.03 | 43.5 | — | |
| PCA←Backbone=CLIP, ViT=L/14, Similarity=⟨I←, I←⟩2026.03 | 42.2 | — | |
| OTIBackbone=SigLIP, ViT=B/16, Similarity=⟨T, I⟩2026.03 | 38.8 | — | |
| OTIBackbone=CLIP, ViT=L/14, Similarity=⟨T, I⟩2026.03 | 38.2 | — | |
| OriginalBackbone=SigLIP2, ViT=B/16, Similarity=⟨I, I⟩2026.03 | 37.9 | — | |
| OriginalBackbone=SigLIP, ViT=B/16, Similarity=⟨I, I⟩2026.03 | 35.8 | — | |
| OriginalBackbone=CLIP, ViT=L/14, Similarity=⟨I, I⟩2026.03 | 33 | — | |
| PCA←Backbone=CLIP, ViT=B/32, Similarity=⟨I←, I←⟩2026.03 | 28.6 | — | |
| OTIBackbone=CLIP, ViT=B/32, Similarity=⟨T, I⟩2026.03 | 23.8 | — | |
| OriginalBackbone=CLIP, ViT=B/32, Similarity=⟨I, I⟩2026.03 | 21.6 | — | |
| PCA←Backbone=SLIP, ViT=B/16, Similarity=⟨I←, I←⟩2026.03 | 19.1 | — | |
| OriginalBackbone=SLIP, ViT=B/16, Similarity=⟨I, I⟩2026.03 | 15.4 | — | |
| OTIBackbone=SLIP, ViT=B/16, Similarity=⟨T, I⟩2026.03 | 14.8 | — | |
| Gemini Embedding 2Modality=V/A/I/T2026.05 | — | 83.6 | |
| multimodalembedding@001Modality=I/T2026.05 | — | 71.8 |