Binary Classification on NICO++ Bike vs. Motorbike
90.2Accuracy (AVG)BAP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| BAPBackbone=SigLIP 2 ViT-B/16, Evaluation Protocol=Linear Probing2026.05 | 90.2 | 74.1 | |
| Data-Matched ControlBackbone=SigLIP 2 ViT-B/16, Evaluation Protocol=Trained Evaluation2026.05 | 88.6 | 64.3 | |
| Data-Matched ControlBackbone=CLIP ConvNeXt-W, Evaluation Protocol=Linear Probing2026.05 | 87.2 | 70.1 | |
| BAP (LP)Backbone=CLIP ConvNeXt-W, Evaluation Protocol=Linear Probing2026.05 | 85.9 | 80.3 | |
| Native Backbone (LP)Backbone=CLIP ConvNeXt-W, Evaluation Protocol=Linear Probing2026.05 | 83.1 | 55.3 | |
| Native BackboneBackbone=SigLIP 2 ViT-B/16, Evaluation Protocol=Zero-Shot2026.05 | 81.4 | 49.2 | |
| Native BackboneBackbone=SigLIP 2 ViT-B/16, Evaluation Protocol=Linear Probing2026.05 | 78.3 | 24.9 | |
| BAPBackbone=SigLIP 2 ViT-B/16, Evaluation Protocol=Zero-Shot2026.05 | 77.9 | 60.4 | |
| DIALBackbone=SigLIP 2 ViT-B/16, Evaluation Protocol=Zero-Shot2026.05 | 77 | 52.7 | |
| BAP (ZS)Backbone=CLIP ConvNeXt-W, Evaluation Protocol=Zero-shot2026.05 | 76.7 | 69.8 | |
| Native Backbone (ZS)Backbone=CLIP ConvNeXt-W, Evaluation Protocol=Zero-shot2026.05 | 75.8 | 60.1 | |
| DIALBackbone=CLIP ConvNeXt-W, Evaluation Protocol=Zero-shot2026.05 | 73.9 | 64 |