Research map / Vision-language and multimodal / Human and embodied modeling
Coarse-to-Fine Vision-Language Pre-training with Fusion in the Backbone
Zi-Yi Dou, Aishwarya Kamath, Zhe Gan, Pengchuan Zhang, Jianfeng Wang, Linjie Li, Zicheng Liu, Ce Liu, Yann LeCun, Nanyun Peng, Jianfeng Gao, Lijuan Wang
Read the paperCheck your idea against 137,217 papers
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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add gated co-attention rather than using ungated co-attention with 6 fusion layers (ablation) | FIBER · VQAv2 test-dev | VQA score: 68.13 -> 71.97 | Improved |
| Use 6 gated co-attention layers rather than 6 merged-attention layers (ablation) | FIBER · VQAv2 test-dev | VQA score: 70.67 -> 71.97 | Improved |
| Add MLM to coarse-grained pre-training (ablation) | FIBER · VQAv2 test-dev | VQA score: 67.45 -> 74.49 | Improved |
| Add MLM to coarse-grained pre-training (ablation) | FIBER · Flickr30k retrieval | IR@1: 75.20 -> 73.58 | No clear effect |
| Use coarse-grained image-text pre-training before fine-grained training (ablation) | FIBER · RefCOCO+ Val | accuracy: 85.04 -> 85.74 | Improved |
| Use coarse-grained image-text pre-training before fine-grained training (ablation) | FIBER · COCO detection zero-shot | AP: 48.9 -> 49.3 | Improved |
| Continue pre-training from the coarse checkpoint with fine-grained image-text-box data before VQA fine-tuning (ablation) | FIBER · VQAv2 test-dev | VQA score: 78.55 -> 74.3 | Hurt |
| Use seq2seq rather than ladder architecture for captioning (ablation) | FIBER · COCO captioning | CIDEr: 127.5 -> 128.4 | Improved |
| Replace RoBERTa with BERT (ablation) | FIBER without VLP · VQAv2 | VQA score: 71.97 -> 71.86 | Hurt |
| Replace Swin with CLIP-ViT (ablation) | FIBER without VLP · VQAv2 | VQA score: 71.97 -> 71.37 | Hurt |
| Ensemble the ITC dual encoder and ITM fusion encoder | FIBER · Flickr30k retrieval | IR@1: 81.44 -> 90.96 | Improved |
| Add GOLD captioning training | FIBER-B · COCO captioning | CIDEr: 128.4 -> 133.6 | Improved |
Closest papers by meaning
- LoMo: Local Modality Substitution for Deeper Vision-Language Fusion (NeurIPS 2026 · similarity 0.90)
- VILA: On Pre-training for Visual Language Models (CVPR 2024 · similarity 0.90)
- MoDA: Modulation Adapter for Fine-Grained Visual Understanding in Instructional MLLMs (ICML 2026 · similarity 0.89)
- Probing Visual Language Priors in VLMs (ICML 2025 · similarity 0.89)
- Revisiting the Role of Language Priors in Vision-Language Models (ICML 2024 · similarity 0.89)
- Prismatic VLMs: Investigating the Design Space of Visually-Conditioned Language Models (ICML 2024 · similarity 0.89)
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.
