Multimodal Sentiment Analysis on CMU-MOSEI Unaligned
49.7Accuracy (7-class)Full modality (ours)
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Full modality (ours)Modality=Full (complete modality supervision)2023.12 | 49.7 | 79.8 | 80.1 | 0.641 | 0.681 | |
| OursModality=Language and vision (with reconstructed audio)2023.12 | 49.6 | 79.4 | 79.5 | 0.646 | 0.648 | |
| CTC + MCTNAlignment method=CTC, Modality=Full2023.12 | 48.2 | 79.3 | 79.7 | 0.631 | 0.645 | |
| Language and vision (ours)Modality=Language and vision2023.12 | 47.9 | 78.2 | 79.1 | 0.649 | 0.627 | |
| CTC + RAVENAlignment method=CTC, Modality=Full2023.12 | 45.5 | 75.4 | 75.7 | 0.664 | 0.599 | |
| Language only (ours)Modality=Language only2023.12 | 45.4 | 76.9 | 77.1 | 0.66 | 0.623 | |
| Vision only (ours)Modality=Vision only2023.12 | 42.1 | 65.7 | 68.4 | 0.741 | 0.339 |