Research map / 3D vision / Face and shape
FastCAD: Real-Time CAD Retrieval and Alignment from Scans and Videos
Florian Maximilian Langer, Jihong Ju, Georgi Dikov, Gerhard Reitmayr, Mohsen Ghafoorian
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 |
|---|---|---|---|
| predict front-facing side with a separate classification head instead of encoding orientation in embeddings (ablation) | FastCAD · Scan2CAD (scan) | instance alignment accuracy: 56.2 -> 60.1 | Improved |
| use symmetry-aware front-facing targets (ablation) | FastCAD · Scan2CAD (scan) | instance alignment accuracy: 60.1 -> 61.7 | Improved |
| distill CAD embeddings into the detector instead of cropping with predicted bbox and re-encoding (ablation) | FastCAD · Scan2CAD reconstruction accuracy (scan) | reconstruction accuracy: 15.6 -> 41.7 | Improved |
| distill embeddings versus two-step retrieval even with nearest GT bbox (ablation) | FastCAD · Scan2CAD shape accuracy (scan) | shape accuracy: 78.1 -> 83.1 | Improved |
| add Chamfer regression and segmentation auxiliary tasks to encoder training (ablation) | FastCAD encoder · Scan2CAD reconstruction accuracy (scan) | reconstruction accuracy: 38.3 -> 41.7 | Improved |
| add auxiliary tasks to encoder training on instance alignment accuracy (ablation) | FastCAD encoder · Scan2CAD (scan) | instance alignment accuracy: 62.3 -> 61.7 | No clear effect |
| use Perceiver encoder instead of PointNet++ (ablation) | FastCAD encoder · Scan2CAD reconstruction accuracy (scan) | reconstruction accuracy: 29.6 -> 38.3 | Improved |
| remove colour from scan input (ablation) | FastCAD · Scan2CAD (scan) | instance alignment accuracy: 61.7 -> 60.4 | No clear effect |
| reduce input points from 50K to 5K (ablation) | FastCAD · Scan2CAD (scan) | instance alignment accuracy: 61.7 -> 51.4 | Hurt |
| use FastCAD on reconstructed point cloud from video versus frame-based methods | FastCAD (Video) · Scan2CAD (video) | instance alignment accuracy: 43.0 -> 48.2 | Improved |
| FastCAD versus ScanNotate on alignment (ScanNotate starts from GT poses) | FastCAD (Scan) · Scan2CAD (scan) | alignment accuracy: 78.2 -> 61.7 | Hurt |
| display class with inconsistent CAD orientation in ShapeNet | FastCAD · Scan2CAD per-class (scan) | display alignment accuracy: 24.1 vs mean 52.8 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Vid2CAD — Vid2CAD: CAD Model Alignment Using Multi-View Constraints From Videos (TPAMI 2022) · better on 4 of 4 shared results
- RayTran — RayTran: 3D Pose Estimation and Shape Reconstruction of Multiple Objects from Videos with Ray-Traced Transformers (ECCV 2022) · 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
- Scan2CAD: Learning CAD Model Alignment in RGB-D Scans (CVPR 2019 · similarity 0.91)
- VideoCAD: A Dataset and Model for Learning Long‑Horizon 3D CAD UI Interactions from Video (NeurIPS 2025 · similarity 0.88)
- cadrille: Multi-modal CAD Reconstruction with Reinforcement Learning (ICLR 2026 · similarity 0.87)
- Point2CAD: Reverse Engineering CAD Models from 3D Point Clouds (CVPR 2024 · similarity 0.87)
- CADCrafter: Generating Computer-Aided Design Models from Unconstrained Images (CVPR 2025 · similarity 0.86)
- From 2D CAD Drawings to 3D Parametric Models: A Vision-Language Approach (AAAI 2025 · similarity 0.86)
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.
