Kernekoncepter
FastCAD enables real-time CAD retrieval and alignment, outperforming existing methods while significantly reducing inference time.
Resumé
FastCAD proposes a real-time method for CAD retrieval and alignment.
The system achieves high-quality shape retrievals by learning CAD embeddings in a contrastive learning framework.
FastCAD accelerates the inference time by a factor of 50 compared to other methods operating on RGB-D scans.
The approach collaborates seamlessly with online 3D reconstruction techniques, enabling real-time generation of precise CAD model-based reconstructions from videos at 10 FPS.
Key contributions include novel methods for CAD model-based reconstruction, efficient system for predicting CAD retrievals and alignments, state-of-the-art alignment accuracy on Scan2CAD benchmark, and new evaluation metrics assessing the quality of retrieved shapes.
Statistik
Our single-stage method accelerates the inference time by a factor of 50 compared to other methods operating on RGB-D scans.
Doing so, we significantly improve the Scan2CAD alignment accuracy in the video setting from 43.0% to 48.2% and the reconstruction accuracy from 22.9% to 29.6%.