Research map / 3D vision / Human pose estimation
Deep Autoencoder for Combined Human Pose Estimation and Body Model Upscaling
Matthew Trumble, Andrew Gilbert, Adrian Hilton, John Collomosse
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 the decoder and volume reconstruction loss (dual loss) to the pose encoder (ablation) | pose autoencoder · TotalCapture | mean joint error (mm): 85.4 -> 38.6 | Improved |
| increase upscaling from x1 to x2 | AutoEnc-LSTM · TotalCapture | mean joint error (mm): 38.6 -> 34.1 | Improved |
| increase upscaling from x2 to x4 | AutoEnc-LSTM · TotalCapture | mean joint error (mm): 34.1 -> 35.5 | No clear effect |
| increase upscaling from x2 to x4 | AutoEnc · TotalCapture | voxel MSE x10^-3: 0.37 -> 0.56 | No clear effect |
| replace tricubic upsampling with the autoencoder at x2 | AutoEnc-x2 · TotalCapture | voxel MSE x10^-3: 9.88 -> 0.37 | Improved |
| replace tricubic upsampling with the autoencoder at x4 | AutoEnc-x4 · TotalCapture | voxel MSE x10^-3: 10.56 -> 0.56 | Improved |
| use the proposed autoencoder instead of the previous best (Martinez) | proposed method · Human3.6M (S9,S11) mean | mean joint error (mm): 62.9 -> 62.5 | Improved |
| use the proposed autoencoder instead of Trumble's earlier CNN | AutoEnc-x2-LSTM · TotalCapture mean | mean joint error (mm): 70.0 -> 34.1 | Improved |
| evaluate on other subject pairs instead of S9/S11 | proposed method · Human3.6M (S9,S11) | mean joint error (mm): 62.5 -> 70.2 | No clear effect |
| raise or lower the pose loss weight lambda outside the working range (ablation) | dual loss training · Human3.6M (S9,S11) | convergence / latent pose quality | Hurt |
Closest papers by meaning
- BodyNet: Volumetric Inference of 3D Human Body Shapes (ECCV 2018 · similarity 0.90)
- 3D Human Pose Estimation From a Single Image via Distance Matrix Regression (CVPR 2017 · similarity 0.89)
- Unite the People: Closing the Loop Between 3D and 2D Human Representations (CVPR 2017 · similarity 0.89)
- Neural Descent for Visual 3D Human Pose and Shape (CVPR 2021 · similarity 0.88)
- Capturing the Motion of Every Joint: 3D Human Pose and Shape Estimation with Independent Tokens (ICLR 2023 · similarity 0.88)
- Hierarchical Kinematic Probability Distributions for 3D Human Shape and Pose Estimation From Images in the Wild (ICCV 2021 · similarity 0.88)
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.
