Multi-modal Continual Learning on VQAv2, ImageNet, Flickr30k, OCRVQA, RefCOCO, HQEdit Sequence
62.72VQAv2 AccuracyUpper-Bound
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Upper-BoundVenue=-2026.02 | 62.72 | 47.88 | 31.24 | 42.84 | 47.68 | 87.49 | 53.31 | — | — | |
| MAGEVenue=-, Evaluation Stage=After final task2026.02 | 45.42 | 68.02 | 31.44 | 27.28 | 34.1 | 91.19 | 49.58 | 12.26 | 59.79 | |
| EWCVenue=PNAS’17, Evaluation Stage=After final task2026.02 | 40.44 | 61.07 | 30.49 | 18.64 | 40.08 | 91.3 | 47 | 15.85 | 60.21 | |
| RegLoRAVenue=ICML’25, Evaluation Stage=After final task2026.02 | 40.34 | 61.1 | 30.42 | 25.72 | 35.18 | 91.6 | 47.39 | 15.05 | 59.94 | |
| PGPVenue=ICLR’24, Evaluation Stage=After final task2026.02 | 38.08 | 74.14 | 30.82 | 18.08 | 33.64 | 91.43 | 47.7 | 13.54 | 58.97 | |
| CIAVenue=ICML’25, Evaluation Stage=After final task2026.02 | 28.4 | 68.91 | 30.49 | 31 | 37.78 | 91.36 | 47.99 | 14.89 | 60.4 | |
| Zero-ShotVenue=-2026.02 | 26.64 | 28.38 | 19.76 | 13.78 | 25.54 | 58.76 | 28.81 | — | — | |
| MoELoRAVenue=NIPS’24, Evaluation Stage=After final task2026.02 | 25.32 | 63.57 | 30.59 | 16.04 | 36.32 | 91.54 | 43.9 | 19.08 | 59.8 | |
| LAEVenue=ICCV’23, Evaluation Stage=After final task2026.02 | 21.56 | 70.71 | 30.63 | 28.54 | 28.68 | 91.39 | 45.25 | 16.29 | 58.82 | |
| LoRAVenue=ICLR’22, Evaluation Stage=After final task2026.02 | 18.4 | 61.01 | 30.6 | 19.38 | 32.7 | 91.27 | 42.23 | 20.36 | 59.19 |