Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Learning of Depth Inference for Multi-View Stereo
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 depth-hypothesis probability volumes for image synthesis instead of direct depth warping (ablation) | Unsupervised CVP-MVSNet training · DTU | f-score: 75.29% -> 77.06% | Helped |
| Add high-resolution pseudo-label refinement (ablation) | Self-supervised CVP-MVSNet · DTU | Overall: 0.378 -> 0.372 | Helped |
| Add multi-view pseudo-label fusion to high-resolution refinement and filtering (ablation) | Self-supervised CVP-MVSNet · DTU | Overall: 0.372 -> 0.363 | Helped |
| Add high-resolution refinement to low-resolution inference, multi-view fusion, and filtering (ablation) | Self-supervised CVP-MVSNet · DTU | Overall: 0.376 -> 0.363 | Helped |
| Iteratively self-train using refined pseudo labels (ablation) | Self-supervised CVP-MVSNet · DTU | f-score: 77.06% -> 88.42% | Helped |
| Continue self-training beyond the second iteration (ablation) | Self-supervised CVP-MVSNet · DTU | f-score: 88.42% -> 88.49% -> 88.47% | No clear effect |
| Replace unsupervised initialization with OpenMVS-generated pseudo labels (ablation) | Self-supervised CVP-MVSNet · DTU | f-score: 88.49% -> 88.02% | No clear effect |
| Use self-supervised rather than unsupervised training with refined pseudo-label self-training | CVP-MVSNet · DTU | Overall (mm): 0.583 -> 0.363 | Helped |
| Train on Tanks and Temples rather than only transfer the model trained on DTU | Self-supervised CVP-MVSNet · Tanks and Temples intermediate set | Mean f-score: 46.71 -> 56.54 | Helped |
| Train without ground-truth depth instead of supervised training | CVP-MVSNet · DTU | f-score: 88.63% -> 88.42% | Hurt |
Closest papers by meaning
- The Temporal Opportunist: Self-Supervised Multi-Frame Monocular Depth (CVPR 2021 · similarity 0.92)
- Self-Supervised Monocular Trained Depth Estimation Using Self-Attention and Discrete Disparity Volume (CVPR 2020 · similarity 0.91)
- Exploiting Pseudo Labels in a Self-Supervised Learning Framework for Improved Monocular Depth Estimation (CVPR 2022 · similarity 0.90)
- Self-supervised surround-view depth estimation with volumetric feature fusion (NeurIPS 2022 · similarity 0.90)
- On the Uncertainty of Self-Supervised Monocular Depth Estimation (CVPR 2020 · similarity 0.90)
- DualNet: Robust Self-Supervised Stereo Matching with Pseudo-Label Supervision (AAAI 2025 · similarity 0.89)
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.