Multimodal Classification on Multimodal MNIST
93.2AccuracyOurs (full model)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Ours (full model)2025.10 | 93.2 | 91.6 | 78.1 | |
| Cross-Modal Attn.2025.10 | 92.4 | 90.8 | 74.3 | |
| Early FusionFusion Strategy=concat.2025.10 | 90.7 | 87.8 | 68.5 | |
| Fusion onlyXAI=no XAI2025.10 | 90.1 | 87.2 | 0 | |
| Text-onlyBackbone=BERT-base2025.10 | 88.5 | 85.2 | 63.9 | |
| Visual-onlyBackbone=ResNet-502025.10 | 87.3 | 84.6 | 61.2 |