COG

KO | EN

gitlite — search

#javascript #hacktoberfest #react #ai #go #golang #android #machine-learning #docker #java #windows #awesome-list #vue #pytorch

(★ 9)

No description available.

COG Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# COG: Confidence-aware Optimal Geometric Correspondence for Unsupervised Single-reference Novel Object Pose Estimation (CVPR2026)

This repo provides the implementation of COG, an unsupervised model for single-reference novel object poes estimation.
Available at [arxiv](https://arxiv.org/abs/2603.00493).
![COG model overview](https://raw.githubusercontent.com/YC-Che/COG/master/assert/cog.png)

## Dependencies

```bash
conda create --name cog python=3.10.12
conda activate cog
pip install torch==2.2.0 torchvision==0.17.0 --index-url https://download.pytorch.org/whl/cu118  # use the correct version of cuda for your system
pip install mmcv==2.2.0 -f https://download.openmmlab.com/mmcv/dist/cu118/torch2.2/index.html # change cuda version if necessary

# Other required packages
pip install -r requirements.txt

# install bop toolkit for evaluation
cd third_party/bop_toolkit
python setup.py install
cd -

# build pointnet2 extension
cd core/COG/model/pointnet2/
pip install -e .
cd -

# Install Detectron2 & PyTorch3D
pip install "git+https://github.com/facebookresearch/detectron2.git"
pip install "git+https://github.com/facebookresearch/pytorch3d.git"

# Create output directory
mkdir output
```

## Datasets

Please refer to [UnoPose](https://github.com/shanice-l/UNOPose) for instructions on downloading the dataset files.

## Checkpoints

**Download DINOv2 checkpoint:** Download `'timm_vit_base_patch14_reg4_dinov2_lvd142m.pth'` and place it in `checkpoints/`:

```bash
# export HF_ENDPOINT="https://hf-mirror.com"  # if huggingface.co is not accessible
python core/COG/scripts/download_and_save_dinov2_ckpt.py
```

**Download COG checkpoint:** Download it from [model weights](https://drive.google.com/file/d/12Y6DRl3y39GxLL2rQJwM4ieWRwZUnk1c/view?usp=sharing).

## Reproduce the results

**Eval**

```bash
./core/COG/save_cog.sh configs/main_cfg.py <gpu_ids> <ckpt_path> (other args)
```

**Notice:** To change the evaluation dataset, `ref_targets_name`, `eval_dataset_name`, and `detetion_path` in the config file need to be replaced. Please search for these parameters in `configs/main_cfg.py`.

**Training**

```bash
./core/COG/dp_train_cog.sh configs/main_cfg.py <gpu_ids> (other args)
```

## Citation

If you find this repo useful in your research, please consider citing:

```
@InProceedings{che2026cog,
      title={COG: Confidence-aware Optimal Geometric Correspondence for Unsupervised Single-reference Novel Object Pose Estimation}, 
      author={Yuchen Che and Jingtu Wu and Hao Zheng and Asako Kanezaki},
      booktitle={CVPR},
      year={2026}
}
```

## Acknowledgments

This work is based on [UnoPose](https://github.com/shanice-l/UNOPose). We thank the authors for their insightful work.

COG

COG Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

COG

COG Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README