Using a pipeline grounded on the Zero to CAD dataset and outputting into CadQuery, the results came back significantly better than I expected. To push it, I tasked the agent with generating two ...

From the left, these are the input CAD image, the ground truth 3D shape, the output of the small-scale data configuration v1, and the output of the large-scale data configuration v2. It is able to ...

AI-native 3D modeling for VS Code: An MCP-powered CAD system using Open CASCADE (OCCT) to let agents see, debug, and build geometry - Taika-3D-Oy/mcpcad ...

Scores of different models across the three tasks in P3D-Bench. The Score is the average of the four bucket scores (Geometry, Topology, Judge, Part), rescaled to 0–100. Multimodal large language ...