Few-shot Image Classification on Aves
69.4AccuracyPOC
Evaluation results
| Method | Links | |
|---|---|---|
| POCbase_method=Few-shot FT, shots=16, strategy=finetuning, variant=POC2025.12 | 69.4 | |
| POCbase_method=CLAP, shots=8, strategy=linear probing, variant=POC2025.12 | 65.9 | |
| POCbase_method=Few-shot FT, shots=4, strategy=finetuning, variant=POC2025.12 | 63 | |
| POCbase_method=CLAP, shots=4, strategy=linear probing, variant=POC2025.12 | 62.3 | |
| POCbase_method=Few-shot LP, shots=4, strategy=linear probing, variant=POC2025.12 | 61.2 | |
| POCbase_method=CrossModal LP, shots=4, strategy=linear probing, variant=POC2025.12 | 61.1 | |
| POCbase_method=CLIP-Adapter, shots=4, strategy=adapter learning, variant=POC2025.12 | 60.3 | |
| POCbase_method=Tip-Adapter, shots=4, strategy=adapter learning, variant=POC2025.12 | 58.9 | |
| POCbase_method=FineR, shots=4, strategy=non-learned, variant=POC2025.12 | 58.7 | |
| POCbase_method=MaPLe, shots=4, strategy=prompt learning, variant=POC2025.12 | 58.6 | |
| Few-shot FTshots=16, strategy=finetuning, variant=expert2025.12 | 58.2 | |
| POCbase_method=CoOp, shots=4, strategy=prompt learning, variant=POC, LMM=Qwen-2.5-VL-7B, expert_backbone=OpenCLIP ViT-B/322025.12 | 58 | |
| CLAPshots=8, strategy=linear probing, variant=expert2025.12 | 53.6 | |
| CLAPshots=4, strategy=linear probing, variant=expert2025.12 | 49.9 | |
| CrossModal LPshots=4, strategy=linear probing, variant=expert2025.12 | 47.4 | |
| Few-shot FTshots=4, strategy=finetuning, variant=expert2025.12 | 46.8 | |
| CLIP-Adaptershots=4, strategy=adapter learning, variant=expert2025.12 | 46 | |
| Tip-Adaptershots=4, strategy=adapter learning, variant=expert2025.12 | 45.9 | |
| Few-shot LPshots=4, strategy=linear probing, variant=expert2025.12 | 45 | |
| MaPLeshots=4, strategy=prompt learning, variant=expert2025.12 | 44.2 | |
| FineRshots=4, strategy=non-learned, variant=expert2025.12 | 42.9 | |
| CoOpshots=4, strategy=prompt learning, variant=expert2025.12 | 42.6 |