Leveraging 3D Geometric Priors in 2D Rotation Symmetry Detection (CVPR 2025)

Ahyun Seo, Minsu Cho

[Paper]  |  [Project Page]

This is the official PyTorch implementation of
Leveraging 3D Geometric Priors in 2D Rotation Symmetry Detection, accepted at CVPR 2025.

---

🛠️ Environment Setup

conda create -n rotsymdetr python=3.8.18 pip -y
conda activate rotsymdetr
bash setup.sh
mkdir weights sym_datasets

---

📂 Datasets and Pretrained Weights

• Dataset: DENDI
• Pretrained Weights: Download

Directory structure after setup:

.
├── sym_datasets
│   └── DENDI
├── weights
│   ├── baseline_3d_best.pth
│   └── prior_3d_best.pth
├── projects 
└── setup.sh

---

🚀 Running the Code

🔎 Inference (with pretrained weights)

./tools/dist_test.sh projects/configs/prior_3d.py weights/prior_3d_best.pth $ngpu

🏋️‍♂️ Training

./tools/dist_train.sh projects/configs/prior_3d.py $ngpu

• For the 3D baseline, use baseline_3d.py and baseline_3d_best.pth.

---

🙏 Acknowledgement

Many thanks to the following project for inspiration and support:

• BEVFormer (ECCV 2022) — GitHub

---

📖 Citation

If you find this work helpful, please consider citing:

@inproceedings{seo2025rotsymdetr,
  author    = {Seo, Ahyun and Cho, Minsu},
  title     = {Leveraging 3D Geometric Priors in 2D Rotation Symmetry Detection},
  booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2025}
}