image-sentence alignment on VALSE counting balanced
pairwise accuracy
- 2021-12-14 — ViLBERT 12-in-1: pairwise accuracy 76.7
| Rank | Model | pairwise accuracy | Accuracy (%) | Paper | Code | Year |
|---|---|---|---|---|---|---|
| 11 | ViLBERT | 58.6 | 50.7 | VALSE: A Task-Independent Benchmark for Vision and Language Models Centered on Linguistic Phenomena | heidelberg-nlp/valse | 2021 |
| 12 | GPT2 | 51.6 | – | VALSE: A Task-Independent Benchmark for Vision and Language Models Centered on Linguistic Phenomena | heidelberg-nlp/valse | 2021 |
| 13 | GPT1 | 51.2 | – | VALSE: A Task-Independent Benchmark for Vision and Language Models Centered on Linguistic Phenomena | heidelberg-nlp/valse | 2021 |
| 14 | VisualBERT | 48.2 | 48.3 | VALSE: A Task-Independent Benchmark for Vision and Language Models Centered on Linguistic Phenomena | heidelberg-nlp/valse | 2021 |