Research map / 3D vision / Human pose estimation
A Comprehensive Analysis of Deep Regression
Stéphane Lathuilière, Pablo Mesejo, Xavier Alameda-Pineda, Radu Horaud
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 mirroring augmentation (ablation) | VGG-16 · Biwi | MAE test: [5.49 5.67] -> [3.66 3.79] | Improved |
| add mirroring augmentation (ablation) | ResNet-50 · FLD | MAE test: [2.05 2.13] -> [1.96 2.05] | Improved |
| use layer normalization instead of batch normalization (ablation) | VGG-16 · FLD | MAE test: [2.61 2.76] -> [2.19 2.31] | Improved |
| add batch normalization before regression layer (ablation) | ResNet-50 · Biwi | MAE test: [3.60 3.71] -> [4.59 4.69] | Hurt |
| add batch normalization after FC2 (ablation) | VGG-16 · FLD | MAE test: [3.67 3.90] -> [2.61 2.76] | Improved |
| add dropout in both FC layers (ablation) | VGG-16 · Biwi | MAE test: [4.47 4.60] -> [3.37 3.48] | Improved |
| fine-tune from CB4 instead of CB5 (ablation) | VGG-16 · Biwi | MAE test: [5.13 5.27] -> [3.66 3.79] | Improved |
| regress from FC2 instead of FC1 (ablation) | VGG-16 · Biwi | MAE test: [5.17 5.31] -> [3.66 3.79] | Improved |
| use heatmap target instead of GAP vector (ablation) | VGG-16 · FLD | MAE test: [2.53 2.67] -> [3.08 3.22] | Hurt |
| use heatmap target instead of GAP vector (ablation) | ResNet-50 · Parse | MAE test: [4.86 5.68] -> [18.90 21.32] | Hurt |
| use MAE loss instead of MSE (ablation) | VGG-16 · MPII | MAE test: [8.42 8.78] -> [7.36 7.69] | Improved |
| use MAE or Huber loss instead of MSE (ablation) | VGG-16 · Parse | MAE test | No clear effect |
Closest papers by meaning
- Deep Mixture of Linear Inverse Regressions Applied to Head-Pose Estimation (CVPR 2017 · similarity 0.89)
- Energy-Based Models for Deep Probabilistic Regression (ECCV 2020 · similarity 0.88)
- Nonlinear Regression via Deep Negative Correlation Learning (TPAMI 2019 · similarity 0.88)
- Generalisation in humans and deep neural networks (NeurIPS 2018 · similarity 0.88)
- Progress and Limitations of Deep Networks to Recognize Objects in Unusual Poses (AAAI 2023 · similarity 0.88)
- CNN Filter DB: An Empirical Investigation of Trained Convolutional Filters (CVPR 2022 · similarity 0.87)
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.
