12.2023 - 08.2024 OpenECAD: An Efficient Visual Language Model for Editable 3D-CAD Design
Tutor: Jianqi Shi, Associate Professor, National Trusted Embedded Software Engineering Technology Research Center
Tech stack: Pytorch / Visual Language Model
- OpenECAD is a fine-tuned model based on pre-trained visual language models, designed to enhance CAD modeling by processing 3D design images and generating structured 2D sketches and 3D construction commands. It leverages the visual, logical, and coding capabilities of these models to integrate seamlessly with existing CAD tools' APIs. A new CAD dataset, derived and augmented from public CAD datasets, was created for training the network. This approach addresses the limitations of discrete 3D shape models and the input restrictions of existing CAD-based models.
- DOI of Published Paper on Computers & Graphics: 10.1016/j.cag.2024.104048
12.2021 - 06.2022 Research and application of AI neural network based on natural driving data
Tutor: Xuesong Li, Associate Professor, CyberC3 Intelligent Vehicle Labs
Tech stack: Pytorch / Transformer
- The neural network model preprocesses vehicle sensor data to extract driver characteristics and make driving decisions similar to those of a human driver.
- The model's robustness is verified and enhanced using simulation software.