Research map / Detection and segmentation / CNN features
Transformer Based Line Segment Classifier With Image Context for Real-Time Vanishing Point Detection in Manhattan World
Xin Tong, Xianghua Ying, Yongjie Shi, Ruibin Wang, Jinfa Yang
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 |
|---|---|---|---|
| Use Transformer feature fusion instead of an MLP classifier (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 82.1% -> 98.2% | Improved |
| Add image-context features to line-segment descriptors (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 84.9% -> 98.9% | Improved |
| Add direction features (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 95.4% -> 98.2% | Improved |
| Use 36 direction intervals instead of 18 (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 97.8% -> 98.2% | Improved |
| Add position features (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 84.9% -> 98.2% | Improved |
| Use 16 sampled points instead of 3 (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 98.0% -> 98.2% | Improved |
| Use 32 sampled points instead of 16 (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 98.2% -> 98.2% | No clear effect |
| Use weighted-sum instead of average pooling for ResNet18 line contexts (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 98.7% -> 98.9% | Improved |
| Use weighted-sum instead of average pooling for RefineNet line contexts (ablation) | TLC line-segment classification · SU3 | line classification accuracy: 98.6% -> 99.0% | Improved |
| Train on SU3 rather than ScanNet | TLC · SVVP | AA@5°: 65.0 -> 67.7 | Improved |
| Train on SU3 rather than ScanNet | TLC · YUD | AA@5°: 76.1 -> 77.1 | Improved |
Closest papers by meaning
- VPDETR: End-to-End Vanishing Point DEtection TRansformers (AAAI 2024 · similarity 0.90)
- Improving Transformer Based Line Segment Detection with Matched Predicting and Re-ranking (AAAI 2025 · similarity 0.88)
- End-to-End Real-Time Vanishing Point Detection with Transformer (AAAI 2024 · similarity 0.88)
- Detecting Vanishing Points Using Global Image Context in a Non-Manhattan World (CVPR 2016 · similarity 0.87)
- Convex Relaxation for Robust Vanishing Point Estimation in Manhattan World (CVPR 2025 · similarity 0.85)
- ScaleLSD: Scalable Deep Line Segment Detection Streamlined (CVPR 2025 · 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.
