Research map / 3D vision / Depth and stereo
GigaDepth: Learning Depth from Structured Light with Branching Neural Networks
Simon Schreiberhuber, Jean-Baptiste Weibel, Timothy Patten, Markus Vincze
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 UNet as backbone with 1920/2 regressor instead of own backbone (ablation) | GigaDepth · Synthetic Unity3D structured-light dataset | outlier ratio o(1) (%): 15.97 -> 12.74 | Improved |
| add the branching MLP regressor to the UNet instead of regressing directly (ablation) | UNet · Synthetic Unity3D structured-light dataset | outlier ratio o(1) (%): 33.41 -> 12.74 | Improved |
| use per-line output weights instead of MLP tree (ablation) | UNet · Synthetic Unity3D structured-light dataset | outlier ratio o(1) (%): 33.41 -> 50.23 | Hurt |
| increase output classes from 288 to 1280 (ablation) | GigaDepth · Synthetic Unity3D structured-light dataset | outlier ratio o(1) (%): 17.16 -> 15.72 | Improved |
| increase output classes from 1920 to 2688 (ablation) | GigaDepth · Synthetic Unity3D structured-light dataset | outlier ratio o(1) (%): 15.97 -> 15.93 | No clear effect |
| increase MLP depth from 1 to 2 layers (ablation) | GigaDepth 640 classes · Synthetic Unity3D structured-light dataset | outlier ratio o(0.1) (%): 44.50 -> 44.37 | No clear effect |
| increase MLP depth from 2 to 3 layers (ablation) | GigaDepth 640 classes · Synthetic Unity3D structured-light dataset | outlier ratio o(0.1) (%): 44.37 -> 38.49 | Improved |
| add LCN input channel (ablation) | GigaDepth 1920 classes · Synthetic Unity3D structured-light dataset | outlier ratio o(0.1) (%): 41.27 -> 38.08 | Improved |
| add vertical jitter augmentation of four pixels during training (ablation) | GigaDepth · Synthetic Unity3D structured-light dataset | depth failure under projector misalignment | Improved |
| use branching regressor on CNN features instead of pure CNN regression | GigaDepth · Synthetic Unity3D structured-light dataset | outlier ratio at sub-pixel thresholds | Improved |
Closest papers by meaning
- HyperDepth: Learning Depth From Structured Light Without Matching (CVPR 2016 · similarity 0.91)
- Connecting the Dots: Learning Representations for Active Monocular Depth Estimation (CVPR 2019 · similarity 0.90)
- Learning Depth with Convolutional Spatial Propagation Network (TPAMI 2019 · similarity 0.88)
- UltraStereo: Efficient Learning-Based Matching for Active Stereo Systems (CVPR 2017 · similarity 0.88)
- Bilateral Cyclic Constraint and Adaptive Regularization for Unsupervised Monocular Depth Prediction (CVPR 2019 · similarity 0.88)
- TurboSL: Dense Accurate and Fast 3D by Neural Inverse Structured Light (CVPR 2024 · 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.
