Research map / 3D vision / Human pose estimation
Deep High-Resolution Representation Learning for Visual Recognition
Jingdong Wang, Ke Sun, Tianheng Cheng, Borui Jiang, Chaorui Deng, Yang Sheng Zhao, Dong Liu, Yadong Mu, Mingkui Tan, Xinggang Wang, Wenyu Liu, Bin Xiao
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 across-stage fusion units (ablation) | HRNetV1-W32 · COCO val pose | AP: 70.8 -> 71.9 | Improved |
| add within-stage fusion units on top of across-stage (ablation) | HRNetV1-W32 · COCO val pose | AP: 71.9 -> 73.4 | Improved |
| add across-stage fusion units (ablation) | HRNetV2-W48 · Cityscapes val | mIoU: 74.8 -> 75.4 | Improved |
| add within-stage fusion units on top of across-stage (ablation) | HRNetV2-W48 · Cityscapes val | mIoU: 75.4 -> 76.4 | Improved |
| use strided convolution instead of bilinear downsampling in fusion (ablation) | HRNet · COCO val pose | AP: 72.6 -> 73.4 | Improved |
| use strided convolution instead of bilinear downsampling in fusion (ablation) | HRNet · Cityscapes val | mIoU: 74.2 -> 76.4 | Improved |
| use sum instead of multiplication for fusion (ablation) | HRNet · COCO val pose | AP: 54.7 -> 73.4 | Improved |
| use sum instead of multiplication for fusion (ablation) | HRNet · Cityscapes val | mIoU: 66.0 -> 76.4 | Improved |
| start with high-to-low streams added gradually instead of all four at the beginning (ablation) | HRNetV1-W32 · COCO val pose | AP: 72.5 -> 73.4 | Improved |
| start with high-to-low streams added gradually instead of all four at the beginning (ablation) | HRNetV2-W48 · Cityscapes val | mIoU: 75.7 -> 76.4 | Improved |
| use HRNetV2 (aggregate low-resolution streams) instead of HRNetV1 for pose (ablation) | HRNet-W32 from scratch · COCO val pose | AP: 73.4 -> 73.6 | No clear effect |
| add ImageNet pretraining | HRNetV1-W32 · COCO val pose (256x192) | AP: 73.4 -> 74.4 | Improved |
Closest papers by meaning
- Deep High-Resolution Representation Learning for Human Pose Estimation (CVPR 2019 · similarity 0.92)
- HRFormer: High-Resolution Vision Transformer for Dense Predict (NeurIPS 2021 · similarity 0.90)
- HR-NAS: Searching Efficient High-Resolution Neural Architectures With Lightweight Transformers (CVPR 2021 · similarity 0.90)
- Win-Win: Training High-Resolution Vision Transformers from Two Windows (ICLR 2024 · similarity 0.87)
- No Pains, More Gains: Recycling Sub-Salient Patches for Efficient High-Resolution Image Recognition (CVPR 2025 · similarity 0.87)
- HIIF: Hierarchical Encoding based Implicit Image Function for Continuous Super-resolution (CVPR 2025 · similarity 0.86)
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.
