Research map / 3D vision / Human pose estimation
VLPose: Bridging the Domain Gap in Pose Estimation With Language-Vision Tuning
Jingyao Li, Pengguang Chen, Xuan Ju, Shu Liu, Hong Li Xu, Jiaya Jia
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 text tokens by plain concatenation to the decoder (ablation) | VLPose ViT-S · HumanArt (ViT-S) | AP: +0.09% | Improved |
| add the relation matcher over text and image tokens (ablation) | VLPose ViT-S · HumanArt (ViT-S) | AP: +0.71%; 36.45 -> 37.16 | Improved |
| use [E, T] as matcher key/value rather than T or [E, E.T] (ablation) | relation matcher · HumanArt (ViT-S) | AP: 36.58 and 36.74 -> 37.16 | Improved |
| use BLIP instead of ALBEF as text encoder (ablation) | VLPose · HumanArt (ViT-S) | AP: 36.78 -> 37.16 | Improved |
| use style prompts instead of random prompts (ablation) | VLPose · HumanArt (ViT-S) | AP: 36.74 -> 37.16 | Improved |
| use style prompts instead of fixed prompts (ablation) | VLPose · HumanArt (ViT-S) | AP: 36.76 -> 37.16 | Improved |
| use dual extractor-injector decoder instead of injector decoder (ablation) | VLPose · HumanArt (ViT-S) | AP: 36.93 -> 37.16 | Improved |
| use dual extractor-injector decoder instead of extractor-injector decoder (ablation) | VLPose · HumanArt (ViT-S) | AP: 37.13 -> 37.16 | Improved |
| use VLPose instead of ViTPose with prompt tuning | ViT-H · HumanArt | AP: 40.77 -> 43.03 | Improved |
| use VLPose (restore original weights) instead of prompt-tuned ViTPose | ViT-H · MSCOCO | AP: 75.1 -> 78.84 | Improved |
| use VLPose instead of ViTPose with last-layer tuning | ViT-H · HumanArt | AP: 41.32 -> 43.03 | Improved |
| apply VLPose to Swin-L backbone | SwinPose-L · HumanArt | AP: 33.21 -> 34.17 | Improved |
Closest papers by meaning
- UniPose: A Unified Multimodal Framework for Human Pose Comprehension, Generation and Editing (CVPR 2025 · similarity 0.86)
- ExtPose: Robust and Coherent Pose Estimation by Extending ViTs (ICML 2025 · similarity 0.85)
- A Dual-Augmentor Framework for Domain Generalization in 3D Human Pose Estimation (CVPR 2024 · similarity 0.85)
- PoseLLaVA: Pose Centric Multimodal LLM for Fine-Grained 3D Pose Manipulation (AAAI 2025 · similarity 0.85)
- Probing Visual Language Priors in VLMs (ICML 2025 · similarity 0.85)
- LVLM-Aided Alignment of Task-Specific Vision Models (CVPR 2026 · similarity 0.85)
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.
