MIT Boosts 2D-to-CAD Conversion with GIFT
MIT News AI · rss · 2026-07-16
Researchers from MIT and other institutions have introduced GIFT (Geometric Inference Feedback Tuning), a system designed to improve the ability of vision-language models to automatically convert 2D designs into CAD programs.
Instead of relying on simple random data augmentation, the core idea is to first assess the model's strengths and weaknesses on the task. It then uses the model's own "almost correct" outputs to generate new training data, combining both errors and successful answers into a highly targeted dataset. This eliminates the need for manual error correction and enables static pre-trained models to achieve better results through inference-time scaling.
Studies show that GIFT outperforms various baseline methods in accuracy while using only about 20% of the compute. The authors note that this approach could be extended to more complex CAD tasks in the future, potentially even improving the manufacturability of 3D designs.
More from Research
- Structural ensembles beat single predictions in TCR:pMHC generalization study — quaidmorris · 2026-07-22
- Structural ensembles, not single predictions, drive robust TCR:pMHC generalization — quaidmorris · 2026-07-22
- enFoldX turns AlphaFold3 ensemble noise into a TCR–peptide–MHC predictor — quaidmorris · 2026-07-22
- RSS launches under OMSF to push structural biology data modeling at scale — MoAlQuraishi · 2026-07-22
- enFoldX tops 8 neoantigen scans and an unseen-peptide benchmark — quaidmorris · 2026-07-22
- enFoldX reaches AUC 0.82 on human VDJdb and transfers to mouse at 0.76 — quaidmorris · 2026-07-22