Research map / 3D vision / Human pose estimation
BodyNet: Volumetric Inference of 3D Human Body Shapes
Gul Varol, Duygu Ceylan, Bryan Russell, Jimei Yang, Ersin Yumer, Ivan Laptev, Cordelia Schmid
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 front-view silhouette re-projection loss (ablation) | BodyNet voxels network · SURREAL (full) | SMPL surface error (mm): 73.6 -> 69.9 | Improved |
| add side-view re-projection loss on top of front view (ablation) | BodyNet voxels network · SURREAL (full) | SMPL surface error (mm): 69.9 -> 68.2 | Improved |
| add end-to-end fine-tuning with intermediate task losses (ablation) | BodyNet with FV+SV re-projection · SURREAL (full) | SMPL surface error (mm): 68.2 -> 65.8 | Improved |
| add intermediate task losses to end-to-end training (ablation) | BodyNet end-to-end FV+SV · SURREAL (full) | SMPL surface error (mm): 70.5 -> 65.8 | Improved |
| add end-to-end fine-tuning without intermediate losses (ablation) | BodyNet FV+SV · SURREAL (full) | SMPL surface error (mm): 68.2 -> 70.5 | Hurt |
| replace shape parameter regression with voxel prediction | BodyNet · SURREAL (full) | SMPL surface error (mm): 74.3 -> 73.6 | Improved |
| replace SMPLify++ fitting with voxel prediction | BodyNet · SURREAL (s10) | SMPL surface error (mm): 86.1 -> 86.3 | Improved |
| add RGB input to segmentation and pose cues (ablation) | Shape network · SURREAL | voxel IOU (%): 56.4 -> 58.1 | Improved |
| use 3D pose input instead of 2D pose (ablation) | Shape network · SURREAL | voxel IOU (%): 47.7 -> 56.3 | Improved |
| add joint multi-task training (ablation) | BodyNet subnetworks · SURREAL | 3D pose mean joint distance (mm): 46.1 -> 40.8 | Improved |
| add joint multi-task training (ablation) | BodyNet segmentation subnetwork · SURREAL | mean parts IOU (%): 59.2 -> 69.2 | Improved |
| add joint multi-task training (ablation) | BodyNet 2D pose subnetwork · SURREAL | PCKh@0.5: 82.7 -> 90.8 | Improved |
Closest papers by meaning
- Accurate 3D Body Shape Regression Using Metric and Semantic Attributes (CVPR 2022 · similarity 0.91)
- 3DPeople: Modeling the Geometry of Dressed Humans (ICCV 2019 · similarity 0.90)
- Keep It SMPL: Automatic Estimation of 3D Human Pose and Shape from a Single Image (ECCV 2016 · similarity 0.90)
- Learning To Regress Bodies From Images Using Differentiable Semantic Rendering (ICCV 2021 · similarity 0.90)
- Learning to Estimate 3D Human Pose and Shape From a Single Color Image (CVPR 2018 · similarity 0.90)
- Deep Autoencoder for Combined Human Pose Estimation and Body Model Upscaling (ECCV 2018 · similarity 0.90)
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.
