Research map / 3D vision / Face and shape
Shape Reconstruction Using Volume Sweeping and Learned Photoconsistency
Vincent Leroy, Jean-Sebastien Franco, Edmond Boyer
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 |
|---|---|---|---|
| use a volumetric receptive field instead of a planar one (ablation) | CNN photoconsistency classifier · DTU surface detection, 5 cameras | classifier accuracy (%): 80.67 -> 82.95 | Improved |
| use a volumetric receptive field instead of a planar one (ablation) | CNN photoconsistency classifier · DTU surface detection, 20 cameras | classifier accuracy (%): 77.87 -> 84.84 | Improved |
| use a volumetric receptive field instead of a planar one (ablation) | CNN photoconsistency classifier · DTU surface detection, 49 cameras | classifier accuracy (%): 75.92 -> 83.45 | Improved |
| use a learned CNN classifier instead of ZNCC (ablation) | volumetric photoconsistency · DTU surface detection, 5 cameras | classifier accuracy (%): 64.98 -> 82.95 | Improved |
| use a learned CNN classifier instead of ZNCC (ablation) | volumetric photoconsistency · DTU surface detection, 49 cameras | classifier accuracy (%): 65.58 -> 83.45 | Improved |
| add more cameras (5 to 49) to the planar classifier (ablation) | planar CNN · DTU surface detection | classifier accuracy (%): 80.67 -> 75.92 | Hurt |
| add more cameras (5 to 49) to the volumetric classifier (ablation) | volumetric CNN · DTU surface detection | classifier accuracy (%): 82.95 -> 83.45 | No clear effect |
| replace average pooling of pairwise features with max-pooling (ablation) | multi-view network · DTU | classifier quality | No clear effect |
| fuse depth maps from all views instead of single depth map | volume sweeping · DTU | median completeness (mm): 0.387 -> 0.296 | Improved |
| fuse depth maps from all views instead of single depth map | volume sweeping · DTU | mean completeness (mm): 1.037 -> 0.532 | Improved |
| volumetric network versus planar patch network of Hartmann et al. (single depth map) | learned photoconsistency · DTU | mean accuracy (mm): 1.563 -> 0.599 | Improved |
| volume sweeping versus SurfaceNet | learned MVS · DTU | mean completeness (mm): 0.892 -> 0.532 | Improved |
Closest papers by meaning
- Learning a Multi-View Stereo Machine (NeurIPS 2017 · similarity 0.89)
- Volume Sweeping: Learning Photoconsistency for Multi-View Shape Reconstruction (IJCV 2020 · similarity 0.89)
- Image-Based 3D Object Reconstruction: State-of-the-Art and Trends in the Deep Learning Era (TPAMI 2019 · similarity 0.89)
- Deep Reflectance Volumes: Relightable Reconstructions from Multi-View Photometric Images (ECCV 2020 · similarity 0.88)
- Deep Volumetric Video From Very Sparse Multi-View Performance Capture (ECCV 2018 · similarity 0.88)
- CoReNet: Coherent 3D Scene Reconstruction from a Single RGB Image (ECCV 2020 · 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.
