Street-view Image Attribute Classification on GSS View Direction
95.28AccuracyCLIP-MHAdapter
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CLIP-MHAdapterParadigm=Parameter-Efficient Adaptation, Trainable Parameters=1.38M2026.02 | 95.28 | 87.95 | 95.19 | 73.19 | |
| CoOpParadigm=Parameter-Efficient Adaptation, Trainable Parameters=8K2026.02 | 92.89 | 80.87 | 92.55 | 56.56 | |
| CLIP-Linear ProbeParadigm=Parameter-Efficient Adaptation, Trainable Parameters=3K2026.02 | 89.51 | 76.96 | 90.06 | 60.65 | |
| ZeroR-TrainerParadigm=Zero-shot Transfer2026.02 | 88.52 | 46.95 | 83.13 | 0 | |
| CLIP-AdapterParadigm=Parameter-Efficient Adaptation, Trainable Parameters=0.52M2026.02 | 87.57 | 76.29 | 88.89 | 69.39 | |
| MaxViTParadigm=Vision Transformer, Trainable Parameters=30.9M2026.02 | 87.38 | 77.99 | 89.06 | 82.35 | |
| Zero-shot CLIPParadigm=Zero-shot Transfer2026.02 | 37.77 | 35.62 | 44.69 | 16.52 |