What Makes Training Multi-Modal Classification Networks Hard?
About
Consider end-to-end training of a multi-modal vs. a single-modal network on a task with multiple input modalities: the multi-modal network receives more information, so it should match or outperform its single-modal counterpart. In our experiments, however, we observe the opposite: the best single-modal network always outperforms the multi-modal network. This observation is consistent across different combinations of modalities and on different tasks and benchmarks. This paper identifies two main causes for this performance drop: first, multi-modal networks are often prone to overfitting due to increased capacity. Second, different modalities overfit and generalize at different rates, so training them jointly with a single optimization strategy is sub-optimal. We address these two problems with a technique we call Gradient Blending, which computes an optimal blend of modalities based on their overfitting behavior. We demonstrate that Gradient Blending outperforms widely-used baselines for avoiding overfitting and achieves state-of-the-art accuracy on various tasks including human action recognition, ego-centric action recognition, and acoustic event detection.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Action Recognition | Kinetics-400 | Top-1 Acc80.4 | 505 | |
| Action Recognition | UCF101 (test) | Accuracy79.3 | 376 | |
| Action Recognition | UCF-101 | Top-1 Acc83.09 | 225 | |
| Audio Classification | AudioSet 20K | -- | 151 | |
| Text-to-Video Retrieval | YouCook2 | Recall@1031.07 | 117 | |
| Audio Classification | ESC50 | Top-1 Acc83.5 | 73 | |
| Audio-Visual Classification | CREMA-D (test) | Accuracy51.75 | 70 | |
| Audio Classification | AudioSet | mAP32.4 | 60 | |
| Emotion Recognition | CREMA-D | -- | 54 | |
| Multimodal Classification | CREMA-D | Accuracy64.65 | 53 |