Visual Question Answering on VQA v2 (Calibration and Error Analysis Metrics)
33.5C2%CVP
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| CVPBackbone=BEiT-3-base, Rel. Time=2.3×2026.06 | 33.5 | 49.7 | 3.12 | 34.9 | 18.2 | 11.7 | |
| 4 samplesBackbone=BEiT-3-base, Rel. Time=4.0×2026.06 | 32.4 | 48.7 | 2.15 | 34.1 | 17 | 10.7 | |
| Mean NetworkBackbone=BEiT-3-base, Rel. Time=1.0×2026.06 | 32.1 | 49.4 | 3.82 | 34.6 | 16.9 | 11.2 | |
| Mean Network + Temp. Sc.Backbone=BEiT-3-base, Rel. Time=1.0×2026.06 | 32.1 | 49.4 | 3.72 | 34.6 | 16.9 | 11.2 | |
| StreamliningBackbone=BEiT-3-base, Rel. Time=2.0×2026.06 | 32.1 | 49.4 | 3.77 | 34.5 | 16.9 | 11.2 | |
| 2 samplesBackbone=BEiT-3-base, Rel. Time=2.0×2026.06 | 31.2 | 47.9 | 2.51 | 33.4 | 16 | 9.3 |