Research map / Point clouds / Point cloud classification
Deep Vanishing Point Detection: Geometric Priors Make Dataset Variations Vanish
Yancong Lin, Ruben Wiersma, Silvia L. Pintea, Klaus Hildebrandt, Elmar Eisemann, Jan C. van Gemert
Check 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 Hough Transform and spherical convolutions instead of the non-learning LSD and sphere-NMS baseline (ablation) | Vanishing point detection on ScanNet-1% · ScanNet-1% | accuracy | Improved |
| Replace classic Canny edge features with learned image semantics (ablation) | Hough Transform and spherical-convolution model · ScanNet-1% | accuracy | Improved |
| Backpropagate through the Hough Transform (ablation) | Hough Transform and spherical-convolution model · ScanNet-1% | accuracy | Improved |
| Add spherical convolutions (ablation) | Hough Transform model · ScanNet-1% | accuracy | Improved |
| Increase Hough angle sampling from 90 to 180 (ablation) | Vanishing point detector · SU3-10% | AA@3°: 77.1 -> 79.3 | Improved |
| Increase Gaussian sphere sampling from 8K to 32K points (ablation) | Vanishing point detector · SU3-10% | AA@3°: 73.3 -> 79.3 | Improved |
| Use multi-scale sampling rather than dense sphere sampling (ablation) | Ours · Manhattan datasets | FPS: 5.5 -> 23.0 | Improved |
| Use multi-scale sampling (ablation) | Ours · ScanNet | AA@10°: 63.7 -> 62.4 | Hurt |
| Remove the Manhattan orthogonality constraint by using unknown focal length (ablation) | Ours · ScanNet | AA@10°: 63.7 -> 59.9 | Hurt |
| Use geometric priors in place of relying on learned dataset-specific patterns | Vanishing point detection · ScanNet reduced-data subsets | AA@10° | Improved |
| Use geometric priors rather than NeurVPS for training with reduced data | Vanishing point detection · SU3 reduced-data subsets | AA@10° | Improved |
| Transfer from SU3 to ScanNet without dataset-specific tuning | Ours versus CONSAC · ScanNet, trained on SU3 | AA@10°: 39.5 vs 27.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- VaPiD — VaPiD: A Rapid Vanishing Point Detector via Learned Optimizers (ICCV 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- NeurVPS: Neural Vanishing Point Scanning via Conic Convolution (NeurIPS 2019 · similarity 0.91)
- Detecting Vanishing Points Using Global Image Context in a Non-Manhattan World (CVPR 2016 · similarity 0.90)
- VPDETR: End-to-End Vanishing Point DEtection TRansformers (AAAI 2024 · similarity 0.87)
- End-to-End Real-Time Vanishing Point Detection with Transformer (AAAI 2024 · similarity 0.87)
- Convex Relaxation for Robust Vanishing Point Estimation in Manhattan World (CVPR 2025 · similarity 0.85)
- DeCoTR: Enhancing Depth Completion with 2D and 3D Attentions (CVPR 2024 · similarity 0.85)
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.
