Research map / Image generation / Text-to-image diffusion
Visual Bridge: Universal Visual Perception Representations Generating
Yilin Gao, Shuguang Dou, Junzhou Li, Zhiheng Yu, Yin Li, Dongsheng Jiang, Shugong Xu
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 |
|---|---|---|---|
| fine-tune task head on top of zero-shot flow output | Visual Bridge · ImageNet-1K | top-1 accuracy: 81.5 -> 82.1 | Improved |
| fine-tune detection head | Visual Bridge · COCO | mAP: 39.2 -> 42.8 | Improved |
| fine-tune segmentation head | Visual Bridge · ADE20K | mIoU: 44.6 -> 46.2 | Improved |
| start flow from random noise instead of DINOv2 tokens (ablation) | Visual Bridge · ImageNet-1K | top-1 accuracy: 81.5 -> 0.1 | Hurt |
| use flow matching instead of one-step distillation (ablation) | Visual Bridge · ImageNet-1K | top-1 accuracy: 81.2 -> 81.5 | Improved |
| use DINOv2 foundation tokens instead of CLIP tokens (ablation) | Visual Bridge · ImageNet-1K | top-1 accuracy: 78.1 -> 81.5 | Improved |
| increase flow sampling steps from 1 to 2 (ablation) | Visual Bridge (DiT-S) · ImageNet | top-1 accuracy: 77.3 -> 77.8 | Improved |
| increase flow sampling steps from 2 to 10 (ablation) | Visual Bridge (DiT-S) · ImageNet | top-1 accuracy: 77.8 -> 77.9 | Improved |
| use zero-shot flow instead of Mask2Former baseline | Visual Bridge · ADE20K | PQ: 39.7 -> 37.9 | Hurt |
| use zero-shot flow instead of DETR baseline | Visual Bridge · COCO | mAP: 41.9 -> 39.2 | Hurt |
Closest papers by meaning
- Scaling Properties of Diffusion Models For Perceptual Tasks (CVPR 2025 · similarity 0.88)
- Perception Encoder: The best visual embeddings are not at the output of the network (NeurIPS 2025 · similarity 0.88)
- Diff-2-in-1: Bridging Generation and Dense Perception with Diffusion Models (ICLR 2025 · similarity 0.87)
- DICEPTION: A Generalist Diffusion Model for Visual Perceptual Tasks (NeurIPS 2025 · similarity 0.87)
- UVU: Improving Multimodal Understanding via Vision-Language Unified Autoregressive Paradigm (CVPR 2026 · similarity 0.87)
- What Matters When Repurposing Diffusion Models for General Dense Perception Tasks? (ICLR 2025 · similarity 0.87)
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.
