Papers with subjectivity issues
Topic and Style-aware Transformer for Multimodal Emotion Recognition (2023.findings-acl)
Copied to clipboard
| Challenge: | Existing studies show that visual modality makes minimal contribution to multimodal emotion recognition due to its high dimensionality. |
| Approach: | They propose to leverage the strong multimodality backbone VATT to project the visual signal to the common space with language and acoustic signals. |
| Outcome: | The proposed model outperforms SOTA results and integrates visual signals and handles subjectivity issues by serving as content "normalization" previous studies show that visual modality makes minimal contribution to the performance of multimodal emotion recognition tasks due to high dimensionality. |