https://github.com/col14m/cadrille
[ICLR2026] cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning
https://github.com/col14m/cadrille
cad cadquery llm pytorch qwen2-vl vlm
Last synced: 26 days ago
JSON representation
[ICLR2026] cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning
- Host: GitHub
- URL: https://github.com/col14m/cadrille
- Owner: col14m
- License: apache-2.0
- Created: 2025-05-27T14:12:55.000Z (about 1 year ago)
- Default Branch: master
- Last Pushed: 2026-02-01T16:04:35.000Z (4 months ago)
- Last Synced: 2026-02-02T01:14:13.507Z (4 months ago)
- Topics: cad, cadquery, llm, pytorch, qwen2-vl, vlm
- Language: Python
- Homepage:
- Size: 521 KB
- Stars: 94
- Watchers: 5
- Forks: 13
- Open Issues: 6
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
## `cadrille`: Multi-modal CAD Reconstruction with Online Reinforcement Learning
**News**:
* :fire: Jan, 2026. `cadrille` is accepted to ICLR 2026.
* :fire: May, 2025. `cadrille` is state-of-the-art in three CAD reconstruction benchmarks: DeepCAD, Fusion360, CC3D.
This repository contains an implementation of `cadrille`, a multi-modal (point clouds / images / text) 3D CAD reconstruction method introduced in our paper:
> **cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning**
> [Maksim Kolodiazhnyi](https://github.com/col14m),
> [Denis Tarasov](https://dt6a.github.io),
> [Dmitrii Zhemchuzhnikov](https://github.com/zhemdi),
> [Alexander Nikulin](https://howuhh.github.io),
> [Ilya Zisman](https://zis.mn),
> [Anna Vorontsova](https://highrut.github.io),
> [Anton Konushin](https://scholar.google.com/citations?user=ZT_k-wMAAAAJ),
> [Vladislav Kurenkov](https://dunnolab.ai),
> [Danila Rukhovich](https://github.com/filaPro)
> https://arxiv.org/abs/2505.22914
### Installation
Install Python packages according to our [Dockerfile](Dockerfile). We support DeepCAD (test), Fusion360 (test), Text2CAD (train / val / test), and CAD-Recode (train, val) datasets. Follow our [instruction](data/README.md) to download and preprocess data.
### Train
To start training run *train.py* script:
```shell
python train.py --mode pc_img --use-text
```
To disable some of the modalities set *--mode* to *img* or *pc*, or disable *--use-text*. We don't provide RL fine-tuning code for now. Alternatively both [SFT](https://huggingface.co/maksimko123/cadrille) and [RL](https://huggingface.co/maksimko123/cadrille-rl) models can be downloaded from :hugs: HuggningFace.
### Inference
To predict CadQuery codes run *test.py* script:
```shell
python test.py --split deepcad_test_mesh --mode pc
```
To run on other datasets and modalities use *--split fusion360_test_mesh* or set *--mode* to *img* or *text*.
### Evaluation
To evaluate IoU, invalidity ratio, and chamfer distance run *evaluate.py* script:
```shell
python evaluate.py
```
### Citation
If you find this work useful for your research, please cite our paper:
```
@article{kolodiazhnyi2025cadrille,
title={cadrille: Multi-modal CAD Reconstruction with Online Reinforcement Learning},
author={Maksim Kolodiazhnyi, Denis Tarasov, Dmitrii Zhemchuzhnikov, Alexander Nikulin, Ilya Zisman, Anna Vorontsova, Anton Konushin, Vladislav Kurenkov, Danila Rukhovich},
journal={arXiv preprint arXiv:2505.22914},
year={2025}
}
```