Visual Reasoning on NLVR2 (test-dev)
79.87AccuracyGIVL (900K)
Evaluation results
| Method | Links | |
|---|---|---|
| GIVL (900K)#Param=112M, Data=3.17M2023.01 | 79.87 | |
| GIVL#Param=112M, Data=3.17M2023.01 | 79.03 | |
| VinVL*#Param=112M, Data=3.17M2023.01 | 78.54 | |
| VILLAData=4M2023.01 | 78.39 | |
| OscarData=4.1M2023.01 | 78.07 | |
| ViLBERT#Param=274M, Data=3.3M2023.01 | 77.4 | |
| ViCHA2023.01 | 77.27 | |
| UNITER#Param=300M, Data=4M2023.01 | 77.18 | |
| PixelBERT#Param=114M2023.01 | 76.73 | |
| SOHO2023.01 | 76.37 | |
| ViLT#Param=87M, Data=4M2023.01 | 75.7 | |
| VLMixerData=4M2023.01 | 75.28 | |
| LXMERT#Param=240M2023.01 | 74.9 | |
| VL-T5#Param=224M2023.01 | 74.6 |