🦁 Zoo3D: Zero-Shot 3D Object Detection at Scene Level 🐼

🐾 News

🦉 Zoo3D is accepted to CVPR 2026!

🦁 We released Zoo3D₀!

⏳🐢 Code will be released soon! Stay tuned

This repository contains an implementation of Zoo3D, a zero-shot indoor 3D object detection method introduced in our paper:

Zoo3D: Zero-Shot 3D Object Detection at Scene Level
Andrey Lemeshko, Bulat Gabdullin, Nikita Drozdov, Anton Konushin, Danila Rukhovich, Maksim Kolodiazhnyi
https://arxiv.org/abs/2511.20253

Data Preparation

Follow steps from data/scannet to prepare data.

Running Zoo3D₀

1. Follow the Further installation steps from MaskClustering to install all required dependencies
2. Move our mask_predict.py script into cropformer

cp mask_predict.py Zoo3D_0/third_party/detectron2/projects/CropFormer/demo_cropformer

3. Clone sam2 repository into third_party folder and install it

cd Zoo3D_0/third_party
git clone git@github.com:facebookresearch/sam2.git
cd sam2
pip install -e .

4. Locate the corresponding configuration file in the configs folder
5. Update the following variables:
   • cropformer_path in the configuration file
   • CUDA_LIST in run.py
6. Run the following command:

cd Zoo3D_0
python run.py --config config_name

Result will be stored at data/prediction/config_name

Metrics

Dataset	Task	mAP@25	mAP@50
Scannet200	point clouds	23.6	16.0
Scannet200	posed images	15.7	7.5
Scannet200	unposed images	8.7	3.1

Predictions Example

Citation

If you find this work useful for your research, please cite our paper:

@article{lemeshko2025zoo3d,
  title={Zoo3D: Zero-Shot 3D Object Detection at Scene Level},
  author={Lemeshko, Andrey and Gabdullin, Bulat and Drozdov, Nikita and Konushin, Anton and Rukhovich Danila and Kolodiazhnyi, Maksim},
  journal={arXiv preprint arXiv:2511.20253},
  year={2025}
}