BridgeTower: Building Bridges Between Encoders in Vision-Language Representation Learning
About
Vision-Language (VL) models with the Two-Tower architecture have dominated visual-language representation learning in recent years. Current VL models either use lightweight uni-modal encoders and learn to extract, align and fuse both modalities simultaneously in a deep cross-modal encoder, or feed the last-layer uni-modal representations from the deep pre-trained uni-modal encoders into the top cross-modal encoder. Both approaches potentially restrict vision-language representation learning and limit model performance. In this paper, we propose BridgeTower, which introduces multiple bridge layers that build a connection between the top layers of uni-modal encoders and each layer of the cross-modal encoder. This enables effective bottom-up cross-modal alignment and fusion between visual and textual representations of different semantic levels of pre-trained uni-modal encoders in the cross-modal encoder. Pre-trained with only 4M images, BridgeTower achieves state-of-the-art performance on various downstream vision-language tasks. In particular, on the VQAv2 test-std set, BridgeTower achieves an accuracy of 78.73%, outperforming the previous state-of-the-art model METER by 1.09% with the same pre-training data and almost negligible additional parameters and computational costs. Notably, when further scaling the model, BridgeTower achieves an accuracy of 81.15%, surpassing models that are pre-trained on orders-of-magnitude larger datasets. Code and checkpoints are available at https://github.com/microsoft/BridgeTower.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Visual Question Answering | VQA v2 (test-dev) | Overall Accuracy78.66 | 664 | |
| Visual Question Answering | VQA v2 (test-std) | Accuracy78.73 | 466 | |
| Natural Language Visual Reasoning | NLVR2 (test-p) | Accuracy83.09 | 327 | |
| Natural Language Visual Reasoning | NLVR2 (dev) | Accuracy81.85 | 288 | |
| Visual Entailment | SNLI-VE (test) | Overall Accuracy81.19 | 197 | |
| Image Retrieval | Flickr30K | R@185.83 | 144 | |
| Text Retrieval | Flickr30K | R@194.73 | 75 | |
| Visual Entailment | SNLI-VE (dev) | Accuracy81.11 | 70 | |
| Text Retrieval | COCO-CFs (test) | R@126.36 | 5 | |
| Image Retrieval | COCO-CFs (test) | Recall@121.44 | 5 |