Image Classification on Galaxy10
87AccuracyVISReg
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VISRegBackbone=ViT-B/16, Evaluation Protocol=Fine-tuning2026.06 | 87 | — | — | |
| DINOBackbone=ViT-B/16, Evaluation Protocol=Fine-tuning2026.06 | 86.6 | — | — | |
| VISReg-Inet22KBackbone=ViT-L/14, Training Scale=large scale datasets, Evaluation protocol=Linear probe2026.06 | 79.82 | — | — | |
| DINOv2-LVD142MBackbone=ViT-L/14, Training Scale=large scale datasets, Evaluation protocol=Linear probe2026.06 | 76.72 | — | — | |
| VISRegBackbone=ViT-L/14, Heuristics=Excluded, Evaluation protocol=Linear probe2026.06 | 76.32 | — | — | |
| VISRegBackbone=ViT-B/16, Heuristics=Excluded, Evaluation protocol=Linear probe2026.06 | 74.01 | — | — | |
| MoCoV3Backbone=ViT-B/16, Heuristics=Included, Evaluation protocol=Linear probe2026.06 | 73.06 | — | — | |
| DINOBackbone=ViT-B/16, Heuristics=Included, Evaluation protocol=Linear probe2026.06 | 72.77 | — | — | |
| iBOTBackbone=ViT-L/16, Heuristics=Included, Evaluation protocol=Linear probe2026.06 | 72.66 | — | — | |
| MAEBackbone=ViT-L/16, Heuristics=Excluded, Evaluation protocol=Linear probe2026.06 | 71.98 | — | — | |
| iBOTBackbone=ViT-B/16, Heuristics=Included, Evaluation protocol=Linear probe2026.06 | 71.65 | — | — | |
| I-JEPABackbone=ViT-H/14, Heuristics=Included, Evaluation protocol=Linear probe2026.06 | 71.31 | — | — | |
| data2vecBackbone=ViT-L/16, Heuristics=Included, Evaluation protocol=Linear probe2026.06 | 65.73 | — | — | |
| VLLM + S2COPE (Ours)Labels=Not Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 64.83 | — | — | |
| VLLM (Unoptimized)Labels=Not Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 54.09 | — | — | |
| LaBoLabels=Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 53.55 | — | — | |
| LF-CBMLabels=Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 53.44 | — | — | |
| TextSpanLabels=Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 24.07 | — | — | |
| DCLIPLabels=Required, Architecture=Qwen3-VL-8B, Evaluation Protocol=Concept-based linear probing2026.06 | 11.75 | — | — | |
| LeJEPABackbone=ResNet-50, Training Epochs=2002026.05 | — | 70.1 | 74.4 | |
| SPHERE-JEPABackbone=ResNet-50, Training Epochs=2002026.05 | — | 69.9 | 74.4 |