AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

BridgeTower: Building Bridges between Encoders in Vision-Language Representation Learning

Xiao Xu, Chenfei Wu, Shachar Rosenman, Vasudev Lal, Wanxiang Che, Nan Duan

AAAI 2023 · Vision-language and multimodal › Cross-modal learning · approach: Contrastive vision-language alignment · 64 citations (OpenAlex)

Read the paper

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Connect uni-modal encoder layers to each cross-modal layer with Add&Norm bridge layersMETER-style VL model · VQAv2 test-stdaccuracy: 78.73% vs 77.64Improved
Connect uni-modal encoder layers to each cross-modal layer with bridge layersMETER-style VL model · Flickr30K 1K testRSUM: 576.6 vs 570.7Improved
Replace Add&Norm with elementwise product for the bridge (ablation)BridgeTower · VQAv2 test-devaccuracy: 75.18 -> 73.41Hurt
Replace Add&Norm with learned interpolation, gated interpolation, or a residual learned projection (ablation)BridgeTower · VQAv2 test-devaccuracy: 75.18; 75.09; 75.13; 75.10No clear effect
Replace Add&Norm with concatenation projection, nonlinear projection, cross-attention, or cross-attention plus FFN (ablation)BridgeTower · VQAv2 test-devaccuracy: 75.18 -> 74.55; 74.26; 73.67; 73.54Hurt
Use 2, 3, 4, or 5 rather than 6 cross-modal layers (ablation)BridgeTower · VQAv2 test-devaccuracy: 75.18; 74.12; 74.36; 75.00; 74.98Hurt
Increase the cross-modal encoder from 6 to 8, 10, or 12 layers (ablation)BridgeTower · VQAv2 test-devaccuracy: 75.18; 75.07; 75.06; 74.94No clear effect
Replace internal bridge-connected layers with external cross-modal layers (ablation)BridgeTower · VQAv2 test-devaccuracy: 75.18 -> 74.04Hurt
Replace internal bridge-connected layers with weighted-sum multi-layer feature fusion (ablation)BridgeTower · Flickr30K 1K testRSUM: 533.8 -> 531.4Hurt
Add VG-QA COCO-only question-answer pairs during fine-tuning (ablation)BridgeTower-base · VQAv2 test-devaccuracy: 78.66 -> 78.77Improved
Add all VG-QA question-answer pairs during fine-tuning (ablation)BridgeTower-base · VQAv2 test-devaccuracy: 78.66 -> 79.10Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy