Research map / Architectures and efficiency / Quantization
Quantized Densely Connected U-Nets for Efficient Landmark Localization
Zhiqiang Tang, Xi Peng, Shijie Geng, Lingfei Wu, Shaoting Zhang, Dimitris Metaxas
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 |
|---|---|---|---|
| replace stacked U-Nets with order-1 densely connected U-Nets (16 U-Nets) | stacked U-Nets · MPII validation | parameters: 50.5M -> 15.9M | Improved |
| replace stacked U-Nets with DU-Net (8 U-Nets) | stacked U-Nets · MPII validation | PCKh: 89.3 -> 89.5 | No clear effect |
| replace stacked U-Nets with DU-Net (4 U-Nets) | stacked U-Nets · MPII validation | PCKh: 88.3 -> 88.2 | No clear effect |
| add order-1 connectivity (order-1 vs order-0) (ablation) | stacked U-Nets · MPII validation | PCKh | Improved |
| raise connectivity order from 1 to 7 (ablation) | order-1 DU-Net(8) · MPII validation (epoch 150) | PCKh: 87.1 -> 86.7 | Hurt |
| use memory-efficient shared-memory implementation | DU-Net order-1 · 12GB GPU batch 16 | max trainable U-Nets: 9 -> 16 | Improved |
| add a second iterative pass (detection supervision) | DU-Net(4) · 300-W full set | NME: 4.01 -> 3.93 | Improved |
| add a second iterative pass (regression supervision) | DU-Net(4) · 300-W full set | NME: 3.34 -> 3.28 | Improved |
| use regression instead of detection supervision | DU-Net(4) · 300-W full set | NME: 4.01 -> 3.34 | Improved |
| add iterative refinement to DU-Net(4) instead of doubling depth to DU-Net(8) | DU-Net(4) · 300-W full set | NME: 3.28 vs 3.26 | No clear effect |
| binarize weights without scaling factor | DU-Net(4) · 300-W full set | NME: 3.38 -> 3.75 | Hurt |
| add float scaling factor alpha to binarized weights (ablation) | BW DU-Net(4) · 300-W full set | NME: 3.75 -> 3.58 | Improved |
Closest papers by meaning
- Towards Efficient U-Nets: A Coupled and Quantized Approach (TPAMI 2019 · similarity 0.95)
- Binarized Convolutional Landmark Localizers for Human Pose Estimation and Face Alignment With Limited Resources (ICCV 2017 · similarity 0.87)
- Hierarchical Binary CNNs for Landmark Localization with Limited Resources (TPAMI 2018 · similarity 0.87)
- VS-Net: Voting With Segmentation for Visual Localization (CVPR 2021 · similarity 0.87)
- Improving Landmark Localization With Semi-Supervised Learning (CVPR 2018 · similarity 0.86)
- HyperPose: Hypernetwork-Infused Camera Pose Localization and an Extended Cambridge Landmarks Dataset (CVPR 2025 · 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.
