Research map / 3D vision / Human pose estimation
Hierarchical Binary CNNs for Landmark Localization with Limited Resources
Adrian Bulat, Georgios Tzimiropoulos
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 original bottleneck with hierarchical parallel multi-scale block (same parameters) | binary HG · MPII val | PCKh: 67.2% -> 72.7% | Improved |
| remove 1x1 convolutions from the multi-scale block (ablation) | binary HG · MPII val | PCKh: 69.3% -> 75.5% | Improved |
| widen the thin 3x3 layer from 128 to 256 channels (ablation) | binary bottleneck · MPII val | PCKh: 67.2% -> 70.7% | Improved |
| add multi-scale filtering (max-pool branch and decomposed 5x5) (ablation) | binary bottleneck · MPII val | PCKh: 67.2% -> 69.3% | Improved |
| add data augmentation (ablation) | binary Ours Final · MPII val | PCKh: 72.1% -> 76% | Improved |
| use sigmoid cross-entropy loss instead of L2 (ablation) | binary Ours Final · MPII val | PCKh: 73.8% -> 76% | Improved |
| use max pooling instead of average pooling (ablation) | binary Ours Final · MPII val | PCKh: 71.9% -> 76% | Improved |
| add ReLU after each convolutional layer (ablation) | binary Ours Final · MPII val | PCKh: 76% -> 78.1% | Improved |
| replace residual blocks in upsampling branches with skip concatenation (improved HG) (ablation) | binary HG with Ours block · MPII val | PCKh: 76% -> 76.6% | Improved |
| stack a second binarized HG network (ablation) | binary HG with Ours block · MPII val | PCKh: 76% -> 79.9% | Improved |
| stack a third binarized HG network (ablation) | binary HG with Ours block · MPII val | PCKh: 79.9% -> 81.3% | Improved |
| increase block cardinality up to 16 (ablation) | binary Ours block · MPII val | PCKh: 1% lower | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HyperFace — HyperFace: A Deep Multi-Task Learning Framework for Face Detection, Landmark Localization, Pose Estimation, and Gender Recognition (TPAMI 2017) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Binarized Convolutional Landmark Localizers for Human Pose Estimation and Face Alignment With Limited Resources (ICCV 2017 · similarity 0.98)
- Facial Landmark Detection with Tweaked Convolutional Neural Networks (TPAMI 2017 · similarity 0.89)
- Adversarial Learning of Structure-Aware Fully Convolutional Networks for Landmark Localization (TPAMI 2019 · similarity 0.88)
- Disentangling 3D Pose in a Dendritic CNN for Unconstrained 2D Face Alignment (CVPR 2018 · similarity 0.88)
- Large Pose 3D Face Reconstruction From a Single Image via Direct Volumetric CNN Regression (ICCV 2017 · similarity 0.87)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · 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.
