Research map / 3D vision / Depth and stereo
ActiveStereoNet: End-to-End Self-Supervised Learning for Active Stereo Systems
Yinda Zhang, Sameh Khamis, Christoph Rhemann, Julien Valentin, Adarsh Kowdle, Vladimir Tankovich, Michael Schoenberg, Shahram Izadi, Thomas Funkhouser, Sean Fanello
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 invalidation mask excluding occluded pixels from the reconstruction loss (ablation) | ActiveStereoNet · Synthetic active stereo test set | percentage of pixels with disparity error below threshold | Improved |
| use WLCN reconstruction loss instead of photometric or VGG perceptual loss (ablation) | ActiveStereoNet · Synthetic and real active stereo | pixels with error smaller than 1 disparity | Improved |
| add adaptive support weight window-based loss instead of single-pixel loss (ablation) | ActiveStereoNet · Real active stereo scenes | thin structure recovery and edge fattening | Improved |
| train self-supervised instead of supervised (ablation) | ActiveStereoNet · Synthetic active stereo test set | pixels with error below 5 disparity vs below 2 pixels | Mixed |
| use full resolution inputs instead of half resolution (ablation) | ActiveStereoNet · Real active stereo scenes | thin structures and boundary detail | Improved |
| use both active and passive illumination instead of either alone (ablation) | ActiveStereoNet · Living room scene | disparity quality | Improved |
| increase training set size from 100 to 10000 images (ablation) | ActiveStereoNet · Real unseen scenes | disparity sharpness: 100 -> 10000 images | Improved |
| use WLCN self-supervised loss instead of supervised loss on SceneFlow-trained passive stereo | Passive stereo network · Kitti 2012 and Kitti 2015 (trained on SceneFlow) | EPE | Improved |
Closest papers by meaning
- ActiveZero: Mixed Domain Learning for Active Stereovision With Zero Annotation (CVPR 2022 · similarity 0.90)
- Active-Passive SimStereo - Benchmarking the Cross-Generalization Capabilities of Deep Learning-based Stereo Methods (NeurIPS 2022 · similarity 0.90)
- GS-ASM: 2DGS-Supervised Active Stereo Matching (CVPR 2026 · similarity 0.89)
- StereoNet: Guided Hierarchical Refinement for Real-Time Edge-Aware Depth Prediction (ECCV 2018 · similarity 0.89)
- ActiveZero++: Mixed Domain Learning Stereo and Confidence-Based Depth Completion With Zero Annotation (TPAMI 2023 · similarity 0.89)
- Active Event-based Stereo Vision (CVPR 2025 · 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.
