Research map / 3D vision / Human pose estimation
Learning to Fuse 2D and 3D Image Cues for Monocular Body Pose Estimation
Bugra Tekin, Pablo Marquez-Neila, Mathieu Salzmann, Pascal Fua
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 confidence map stream to image stream (early fusion) (ablation) | Image-Only regressor · Human3.6m | 3D joint position error (mm): 124.13 -> 76.41 | Improved |
| add image stream to confidence map stream (late fusion) (ablation) | CM-Only regressor · Human3.6m | 3D joint position error (mm): 79.28 -> 74.12 | Improved |
| learn the fusion layer instead of late fusion (ablation) | two-stream network · Human3.6m | 3D joint position error (mm): 74.12 -> 69.73 | Improved |
| learn the fusion layer instead of early fusion (ablation) | two-stream network · Human3.6m | 3D joint position error (mm): 76.41 -> 69.73 | Improved |
| add sharp-fusion regularization term (ablation) | trainable fusion · Human3.6m Eating | 3D joint position error (mm): 68.30 -> 60.17 | Improved |
| add sharp-fusion regularization term (pruning) (ablation) | trainable fusion · Human3.6m Eating | runtime (sec/frame): 0.013 -> 0.006 | Improved |
| pretrain on synthetic 3D pose data before fine-tuning (ablation) | trainable fusion · KTH Multiview Football II | PCP all parts (%): 83.2 -> 95.2 | Improved |
| replace single-stage fully-convolutional 2D detector with stacked hourglass (ablation) | confidence map stream · Human3.6m | 3D pose accuracy | Improved |
| use fusion of 2D confidence maps and image cues instead of prior single-image methods | trainable fusion · Human3.6m (all actions) | 3D joint position error (mm): 107.26 -> 69.73 | Improved |
| use trainable fusion instead of prior methods on HumanEva-I | trainable fusion · HumanEva-I Walking | 3D joint position error (mm): 36.6 -> 24.41 | Improved |
Closest papers by meaning
- Towards Part-aware Monocular 3D Human Pose Estimation: An Architecture Search Approach (ECCV 2020 · similarity 0.91)
- A Simple yet Effective Baseline for 3D Human Pose Estimation (ICCV 2017 · similarity 0.91)
- Generating Multiple Hypotheses for 3D Human Pose Estimation With Mixture Density Network (CVPR 2019 · similarity 0.91)
- Sparseness Meets Deepness: 3D Human Pose Estimation From Monocular Video (CVPR 2016 · similarity 0.91)
- MonoCap: Monocular Human Motion Capture using a CNN Coupled with a Geometric Prior (TPAMI 2018 · similarity 0.90)
- 3D Human Pose Estimation From a Single Image via Distance Matrix Regression (CVPR 2017 · similarity 0.89)
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.
