Class-Incremental Learning on TinyImageNet (10 steps, 100 base classes, test)
80.55Avg AccuracyiGSP
Evaluation results
| Method | Links | ||
|---|---|---|---|
| iGSP2026.05 | 80.55 | 76.35 | |
| MoE-Adapter(++)2026.05 | 80.23 | 76.35 | |
| GIFTBackbone=CLIP2025.03 | 80.2 | 75.51 | |
| ZSCLBackbone=CLIP2025.03 | 78.61 | 71.62 | |
| ZSCL2026.05 | 78.61 | 71.62 | |
| LwF-VRBackbone=CLIP2025.03 | 74.12 | 67.05 | |
| LwF-VR2026.05 | 74.12 | 67.05 | |
| iCaRLBackbone=CLIP2025.03 | 73.48 | 65.97 | |
| iCaRL2026.05 | 73.48 | 65.97 | |
| CLIP Zero-shotBackbone=CLIP, Mode=Zero-shot2025.03 | 69.55 | 65.59 | |
| CLIPEvaluation protocol=zero-shot2026.05 | 69.55 | 65.59 | |
| LwFBackbone=CLIP2025.03 | 57.6 | 44 | |
| LwF2026.05 | 57.6 | 44 | |
| CLIP Fine-tuneBackbone=CLIP, Mode=Fine-tune2025.03 | 57.05 | 41.54 | |
| Fine-tune2026.05 | 57.05 | 41.54 | |
| DyTox2025.03 | 52.26 | 42.79 | |
| DyTox+2026.05 | 52.26 | 42.79 | |
| UCIR2025.03 | 48.58 | 37.29 | |
| UCIR2026.05 | 48.58 | 37.29 | |
| PASS2025.03 | 47.19 | 39.27 | |
| PASS2026.05 | 47.19 | 39.27 | |
| EEIL2025.03 | 45.03 | 34.64 | |
| EEIL2026.05 | 45.03 | 34.64 | |
| MUC2025.03 | 26.67 | 15.33 | |
| MUC2026.05 | 26.67 | 15.33 | |
| EWC2025.03 | 15.82 | 3.79 | |
| EWC2026.05 | 15.82 | 3.79 |