Introduction

In our paper "SHREC'22: Open-Set 3D Object Retrieval", we have released two datasets for open-set 3D object retrieval. The two datasets are generated based on the Modelnet40 dataset. Here, we release the core code for multi-modal data generation including Voxel Modality Generation, Pointcloud Modality Generation, and 24 Images Modality Generation.

The computed fully datasets can be download here. Examlpe code can be found here.

Settings for generating voxel and pointcloud modalities

Create a virtual environment with command

conda create -n shrec22 python=3.8

Then, activate the environment, and install the following libraries with pip

pip install open3d, Pillow, numpy, rich, scipy

Usage

Run the following code. The generated voxel and pointcloud will be stored in the out direction.

python generate_voxel_pointcloud.py

Settings for generating 24 images modality

• blender 3.0

Install the PIL library for the inside python in blender. Change the work direction to the direction of the inside python of blender. Then, run the following command:

pip install Pillow

Usage

Open generate_24_images.blend with blender 3.0, then run the code. Then, rendered 24 images can be found in the out folder.

Evaluation

Given a given distance matrix, you can evaluate its mAP, NDCG, ANMRR, NN with functions in the metrics.py.

Citation

@article{feng2022shrec,
  title={SHREC’22 track: Open-Set 3D Object Retrieval},
  author={Feng, Yifan and Gao, Yue and Zhao, Xibin and Guo, Yandong and Bagewadi, Nihar and Bui, Nhat-Tan and Dao, Hieu and Gangisetty, Shankar and Guan, Ripeng and Han, Xie and others},
  journal={Computers \& Graphics},
  year={2022},
  publisher={Elsevier}
}