Visual Reasoning on NLVR2 v1 (test/val)
83.8AccuracyCVP
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| CVPBackbone=BEiT-3-base, Rel. Time=2.5×2026.06 | 83.8 | 5.24 | 14.6 | 23.5 | 0.36 | — | 1.27 | 0.23 | |
| Mean NetworkBackbone=BEiT-3-base, Rel. Time=0.9×2026.06 | 83.4 | 5.61 | 8.5 | 17.1 | 0.41 | — | 5.92 | 0.24 | |
| + Temp. Sc.Backbone=BEiT-3-base, Rel. Time=0.9×2026.06 | 83.4 | 5.61 | 8.5 | 17.1 | 0.36 | — | 1.21 | 0.23 | |
| StreamliningBackbone=BEiT-3-base, Rel. Time=2.1×2026.06 | 83.4 | 5.66 | 8.2 | 14.4 | 0.37 | — | 1.38 | 0.23 | |
| 4 samplesBackbone=BEiT-3-base, Rel. Time=4.0×2026.06 | 83.1 | 5.55 | 14 | 21.8 | 0.39 | — | 3.93 | 0.24 | |
| 2 samplesBackbone=BEiT-3-base, Rel. Time=2.0×2026.06 | 82.9 | 5.78 | 10.7 | 19.9 | 0.4 | — | 4.48 | 0.24 |