Object reference game on SNARE (val)
91.2Visual AccuracyVLG
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VLG2022.05 | 91.2 | 78.4 | 84.9 | |
| MATCHBackbone=CLIP ViT-B/32, Replication=true2022.05 | 90.6 | 75.7 | 83.2 | |
| LAGORBackbone=CLIP ViT-B/322022.05 | 89.8 | 75.3 | 82.6 | |
| LAGORBackbone=CLIP ViT-B/32, Replication=true2022.05 | 89.8 | 75 | 82.5 | |
| ViLBERTBackbone=ViLBERT, Pre-training=12-in-1 (Lu et al., 2020)2022.05 | 89.5 | 76.6 | 83.1 | |
| MATCHBackbone=CLIP ViT-B/32, Pooling variant=max-pool2022.05 | 89.2 | 75.2 | 82.2 |