Research map / Architectures and efficiency / CNN design
Lite-HRNet: A Lightweight High-Resolution Network
Changqian Yu, Bin Xiao, Changxin Gao, Lu Yuan, Lei Zhang, Nong Sang, Jingdong Wang
Check 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 |
|---|---|---|---|
| Replace Small HRNet blocks with ShuffleNet shuffle blocks (ablation) | Small HRNet · COCO val | AP: 55.2 -> 62.5 | Improved |
| Replace Small HRNet blocks with ShuffleNet shuffle blocks (ablation) | Small HRNet · MPII val | PCKh: 80.2 -> 85.3 | Improved |
| Drop one 1×1 convolution (ablation) | Wider Naive Lite-HRNet-18 · COCO val | AP: 65.7 -> 63.6 | Hurt |
| Drop two 1×1 convolutions (ablation) | Wider Naive Lite-HRNet-18 · COCO val | AP: 65.7 -> 61.3 | Hurt |
| Replace the dropped 1×1 convolutions with conditional channel weighting (ablation) | Wider Naive Lite-HRNet-18 with two 1×1 convolutions dropped · COCO val | AP: 61.3 -> 64.8 | Improved |
| Add spatial weights (ablation) | Lite-HRNet-18 with two 1×1 convolutions dropped · COCO val | AP: 61.3 -> 62.6 | Improved |
| Add multi-resolution weights (ablation) | Lite-HRNet-18 with two 1×1 convolutions dropped · COCO val | AP: 61.3 -> 63.0 | Improved |
| Use both spatial and cross-resolution weights (ablation) | Lite-HRNet-18 · COCO val | AP: 61.3 -> 64.8 | Improved |
| Use two spatial weights or two cross-resolution weights instead of one of each (ablation) | Lite-HRNet-18 · COCO val | AP: −0.3 | Hurt |
| Change the order of spatial and cross-resolution weighting (ablation) | Lite-HRNet-18 · COCO val | AP | No clear effect |
Closest papers by meaning
- Deep High-Resolution Representation Learning for Visual Recognition (TPAMI 2020 · similarity 0.89)
- Lite Pose: Efficient Architecture Design for 2D Human Pose Estimation (CVPR 2022 · similarity 0.88)
- HRFormer: High-Resolution Vision Transformer for Dense Predict (NeurIPS 2021 · similarity 0.87)
- Deep High-Resolution Representation Learning for Human Pose Estimation (CVPR 2019 · similarity 0.87)
- Adversarial PoseNet: A Structure-Aware Convolutional Network for Human Pose Estimation (ICCV 2017 · similarity 0.86)
- HR-Depth: High Resolution Self-Supervised Monocular Depth Estimation (AAAI 2021 · 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.
