Image Classification on BloodMNIST
96.23Top-1 AccuracyNo PE
Evaluation results
| Method | Links | |
|---|---|---|
| No PEPositional Encoding=None2026.03 | 96.23 | |
| ResNet-18Backbone=ResNet-182026.03 | 95.97 | |
| RoPEPositional Encoding=RoPE2026.03 | 95.59 | |
| Learnable PEPositional Encoding=Learnable2026.03 | 94.48 | |
| LF-CBMLabels=Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 87.32 | |
| VLLM + S2COPE (Ours)Labels=Not Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 79.73 | |
| LaBoLabels=Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 62.5 | |
| VLLM (Unoptimized)Labels=Not Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 55.26 | |
| DCLIPLabels=Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 23.71 | |
| TextSpanLabels=Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 11.04 |