- Home >
CAD-SIGNet Scenarios
Full design history recovery from an input point cloud (top-left). CAD-SIGNet user interaction (bottom-left and right).
CAD-SIGNet Contributions
-
Auto-regressive network for CAD language inference given a point cloud, allowing for interactive scenarios of reverse engineering
-
Multi-modal transformer blocks with layer-wise cross-attention between point cloud and CAD language embedding
-
A Sketch instance Guided Attention (SGA) guiding the layer-wise cross-attention to attend on relevant regions of the point cloud for predicting sketches
-
Experimental validation in two different reverse engineering settings:
-
Full CAD history recovery
-
Conditional auto-completion
-
CAD-SIGNet Architecture
Qualitative Results
Design History Recovery from Point Clouds:
Conditional Auto-Completion from User Input and Point Clouds:
[1] Wu, R., et al., DeepCAD, A deep generative network for computer-aided design models. ICCV 2021
[2] Willis, K., et al., Fusion 360 gallery: A dataset and environment for programmatic cad construction from human design sequences. ACM TOG 2021
[3] Cherenkova, K., et al., Pvdeconv: Point-voxel deconvolution for autoencoding cad construction in 3d. ICIP 2020
[5] Xu, X., et al., Skexgen: Autoregressive generation of cad construction sequences with disentangled codebooks. ICML 2022
[6] Xu, X., et al., Hierarchical neural coding for controllable cad model generation. ICML 2023
Watch the Video!
Acknowledgments
The present work is supported by the National Research Fund, Luxembourg under the BRIDGES2021/IS/16849599/FREE-3D and IF/17052459/CASCADES projects and Artec3D.
Citation
If you use this work, please use the following citation:
BibTeX
author={Mohammad Sadil Khan, Elona Dupont, Sk Aziz Ali, Kseniya Cherenkova, Anis Kacem, Djamila Aouada},
title={CAD-SIGNet: CAD Language Inference from Point Cloud using Layerwise Sketch Instance Guided Attention},
booktitle={Proceedings of the International Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2024}
}