Continual Visual Question Answering on VQA v2 (test)
55APJoint
Evaluation Results
| Method | Links | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| JointVenue=-, Backbone=LLaVA-1.5-7B, Method Category=Upper bound2026.06 | 55 | — | — | — | — | — | — | — | — | — | — | 0 | — | |
| ASRVenue=Ours, Backbone=Qwen2.5-VL-7B, Method Category=Ours2026.06 | 54.2 | — | — | — | — | — | — | — | — | — | — | 2 | — | |
| ASRVenue=Ours, Backbone=InternVL3-8B, Method Category=Ours2026.06 | 53.6 | — | — | — | — | — | — | — | — | — | — | 2.1 | — | |
| INDUCEKV-OVProtocol=OV2026.07 | 53.12 | — | — | — | — | — | — | — | — | — | — | 1.85 | — | |
| INDUCEKV-LVProtocol=LV2026.07 | 52.64 | — | — | — | — | — | — | — | — | — | — | 1.7 | — | |
| ASRVenue=Ours, Backbone=LLaVA-1.5-7B, Method Category=Ours2026.06 | 52 | — | — | — | — | — | — | — | — | — | — | 2.4 | — | |
| CL-MoEBackbone=LLaVA-7B2025.03 | 51.34 | 46.5 | 37.18 | 75.22 | 71.39 | 40.9 | 69.54 | 43.66 | 52.68 | 55.55 | 20.74 | -0.02 | — | |
| CL-MoEProtocol=LV2026.07 | 51.34 | — | — | — | — | — | — | — | — | — | — | -0.02 | — | |
| CL-MoEVenue=CVPR’25, Backbone=LLaVA-7B, Method Category=Continual VQA / VLM methods2026.06 | 51.2 | — | — | — | — | — | — | — | — | — | — | 3.8 | — | |
| KeepLoRAVenue=ICLR’26, Backbone=LLaVA-1.5-7B, Method Category=Continual VQA / VLM methods2026.06 | 50.6 | — | — | — | — | — | — | — | — | — | — | 4.5 | — | |
| QUADVenue=ICCV’25, Backbone=T5-style, Method Category=Continual VQA / VLM methods2026.06 | 50.3 | — | — | — | — | — | — | — | — | — | — | 6.1 | — | |
| BCP-MFAVenue=ICCV’25, Backbone=VQA Enc-Dec, Method Category=Continual VQA / VLM methods2026.06 | 49.4 | — | — | — | — | — | — | — | — | — | — | 7.5 | — | |
| VQACLVenue=CVPR’23, Backbone=T5-style, Method Category=Continual VQA / VLM methods2026.06 | 48.5 | — | — | — | — | — | — | — | — | — | — | 9.2 | — | |
| ERVenue=-, Backbone=LLaVA-1.5-7B, Method Category=Replay-based2026.06 | 47.9 | — | — | — | — | — | — | — | — | — | — | 11.1 | — | |
| LwFVenue=ECCV’16, Backbone=LLaVA-1.5-7B, Method Category=Regularization-based2026.06 | 45.8 | — | — | — | — | — | — | — | — | — | — | 14.2 | — | |
| EWCVenue=PNAS’17, Backbone=LLaVA-1.5-7B, Method Category=Regularization-based2026.06 | 45 | — | — | — | — | — | — | — | — | — | — | 15.7 | — | |
| VanillaVenue=-, Backbone=LLaVA-1.5-7B, Method Category=Reference2026.06 | 44.1 | — | — | — | — | — | — | — | — | — | — | 18.3 | — | |
| VQACLBackbone=LLaVA-7B2025.03 | 43.49 | 34.14 | 32.19 | 66.15 | 63 | 33.01 | 60.91 | 34.64 | 38.48 | 47.94 | 24.42 | 9.1 | — | |
| VQACLProtocol=LV2026.07 | 43.49 | — | — | — | — | — | — | — | — | — | — | 9.1 | — | |
| ERBackbone=LLaVA-7B2025.03 | 41.95 | 29.31 | 25.74 | 63.46 | 65.78 | 31.92 | 58.39 | 45.17 | 34.55 | 46.24 | 18.96 | 10.2 | — | |
| ERProtocol=LV2026.07 | 41.95 | — | — | — | — | — | — | — | — | — | — | 10.2 | — | |
| DERBackbone=LLaVA-7B2025.03 | 41.16 | 26.95 | 21.43 | 64.88 | 66.17 | 31.01 | 55.92 | 44.6 | 32.85 | 47.09 | 20.74 | 11.28 | — | |
| DERProtocol=LV2026.07 | 41.16 | — | — | — | — | — | — | — | — | — | — | 11.28 | — | |
| VSBackbone=LLaVA-7B2025.03 | 39.79 | 28.48 | 24.09 | 61.37 | 67.2 | 29.56 | 54.64 | 33.98 | 32.91 | 45.82 | 19.89 | 12.7 | — | |
| VSProtocol=LV2026.07 | 39.79 | — | — | — | — | — | — | — | — | — | — | 12.7 | — | |
| VQACLBackbone=VL-T52025.03 | 38.77 | 20.47 | 28.02 | 62.55 | 68.61 | 29.35 | 50.66 | 44.45 | 26.36 | 44.65 | 12.6 | 2.9 | — | |
| MASBackbone=LLaVA-7B2025.03 | 37.71 | 31.54 | 22.09 | 60.85 | 46.32 | 32.48 | 56.47 | 30.05 | 35.69 | 42.73 | 18.83 | 14.91 | — | |
| MASProtocol=LV2026.07 | 37.71 | — | — | — | — | — | — | — | — | — | — | 14.91 | — | |
| EWCBackbone=LLaVA-7B2025.03 | 37.28 | 28.12 | 23.02 | 61.5 | 61.08 | 26.13 | 54.29 | 23.65 | 32.25 | 44.97 | 17.83 | 15.27 | — | |
| EWCProtocol=LV2026.07 | 37.28 | — | — | — | — | — | — | — | — | — | — | 15.27 | — | |
| ERBackbone=VL-T52025.03 | 36.99 | 18.64 | 21.36 | 61.27 | 64.17 | 30.29 | 52.84 | 43.39 | 23.31 | 42.75 | 11.85 | 4.8 | — | |
| DERBackbone=VL-T52025.03 | 35.35 | 14.55 | 13.83 | 62.88 | 65.16 | 30.96 | 51.19 | 40.51 | 19.04 | 42.87 | 12.55 | 6.58 | — | |
| VSBackbone=VL-T52025.03 | 34.03 | 15.66 | 19.21 | 59.86 | 32.16 | 27.28 | 47.79 | 32.32 | 20.44 | 41.38 | 10.2 | 11.68 | — | |
| VanillaBackbone=LLaVA-7B2025.03 | 32.51 | 19.25 | 14.81 | 54.59 | 56.97 | 24.23 | 46.2 | 27.58 | 26.09 | 36.47 | 18.89 | 20.69 | — | |
| VanillaProtocol=LV2026.07 | 32.51 | — | — | — | — | — | — | — | — | — | — | 20.69 | — | |
| MASBackbone=VL-T52025.03 | 20.56 | 30.81 | 8.07 | 25.5 | 4 | 31.9 | 32.39 | 26.24 | 24.75 | 19.85 | 2.75 | 21.97 | — | |
| EWCBackbone=VL-T52025.03 | 15.77 | 6.73 | 8.43 | 27.22 | 47.1 | 0.14 | 12.4 | 1.76 | 10.98 | 31.05 | 11.85 | 28.38 | — | |
| VanillaBackbone=VL-T52025.03 | 14.49 | 7.39 | 4.94 | 22.29 | 32.3 | 0.71 | 12.14 | 12.1 | 10.69 | 27.29 | 15.1 | 30.15 | — | |
| CL-MoEReplay samples=0.5k per task2025.11 | — | 46.5 | 37.18 | 75.22 | 71.39 | 40.9 | 69.54 | 43.66 | 52.68 | 55.55 | 20.74 | — | 57.27 | |
| DISCOReplay samples=0.5k per task2025.11 | — | 54.48 | 38.6 | 73.98 | 68.22 | 49.23 | 72.65 | 72.91 | 59.62 | 57.61 | 29.95 | — | 63.09 | |
| Dynamic Gradient GuidanceReplay samples=0.5k per task2025.11 | — | 55.55 | 41.03 | 78.67 | 76.12 | 48.33 | 75.62 | 69.2 | 61.19 | 60.35 | 28.11 | — | 65.17 | |
| HiDEReplay samples=0.5k per task2025.11 | — | 49.27 | 33.62 | 72.27 | 69.11 | 43.72 | 70.17 | 65.36 | 55.24 | 56.42 | 25.81 | — | 59.44 | |
| MultitaskBackbone=VL-T52025.03 | — | 42.89 | 38.27 | 75.96 | 73.34 | 38.01 | 66.9 | 56.52 | 47.46 | 53.59 | 22.94 | — | — | |
| MultiTaskReplay samples=0.5k per task2025.11 | — | 55.15 | 41.88 | 80.74 | 75.47 | 49.81 | 75.97 | 73.03 | 61.02 | 60.54 | 29.49 | — | 66.26 | |
| Multitask (Upper Bound)Backbone=LLaVA-7B2025.03 | — | 55.15 | 41.88 | 80.74 | 75.47 | 49.81 | 75.97 | 73.03 | 61.02 | 60.54 | 29.49 | — | — | |
| SEFEReplay samples=0.5k per task2025.11 | — | 50.55 | 39.46 | 78.42 | 75.96 | 48.43 | 72.86 | 70.5 | 58.05 | 58.54 | 29.95 | — | 63.57 |