Visual Commonsense Reasoning on GD-VCR (test)
72.01AccuracyGIVL (1M)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| GIVL (1M)#Param=112M2023.01 | 72.01 | 70.4 | 4.97 | |
| GIVL w/ CLIP#Param=199M2023.01 | 70.43 | 67.25 | 10.2 | |
| GIVL#Param=112M2023.01 | 70.32 | 68.41 | 6.14 | |
| VinVL#Param=112M2023.01 | 70.2 | 66.78 | 11.04 | |
| GIVL W/O LIEC#Param=112M2023.01 | 69.89 | 66.92 | — | |
| GIVL W/O LIKM#Param=112M2023.01 | 69.56 | 65.96 | — | |
| GIVL-B#Param=112M2023.01 | 69.56 | 65.96 | — | |
| VinVL*#Param=112M2023.01 | 69.07 | 66.45 | 8.46 | |
| VILBERT#Param=274M2023.01 | 59.99 | — | 7.28 | |
| VisualBERT#Param=135M2023.01 | 53.95 | — | 10.42 |