ShapeNetV2 Symmetry Axis Dataset

1. Dataset Description

This dataset is an extension of the ShapeNetV2 dataset, enhanced with symmetry axis annotations for each 3D model. These symmetry axes are designed to support tasks such as pose estimation, object recognition, and 3D model analysis by providing detailed symmetry information.

Key Features:

Objects: 3D models from the ShapeNetV2 dataset, covering various categories. Symmetry Annotations: Each object is annotated with one or more symmetry axes, covering discrete, continuous, texture, and geometric symmetries. Applications: Ideal for use in pose estimation tasks, symmetry-aware machine learning models, and 3D object analysis.

Current Status:

Subset Selection: We have screened approximately 7000 objects from the ShapeNetV2 dataset (out of a total of 55,000 objects). Symmetry Annotations: Only a portion of these 7000 objects currently includes symmetry axis annotations. The remaining models will be annotated and uploaded over time. Future Work: The symmetry axis annotations for the rest of the selected models are expected to be completed by the end of 2024.

2. Data Source

This dataset is built upon the ShapeNetV2 dataset, which contains richly annotated 3D models across various object categories. The symmetry axis data has been manually or algorithmically added to each model.

Original ShapeNetV2 dataset can be accessed here: https://www.paris.inria.fr/archive_ylabbeprojectsdata/megapose/tars/shapenetcorev2.zip

3. Dataset Structure

The dataset is organized as follows:

Models: 3D models from ShapeNetV2 in .obj format. Symmetry Axes: A JSON file for each object containing symmetry axis data, including discrete, and continuous symmetry information.

The JSON file is organized based on the BOP format: https://github.com/thodan/bop_toolkit

4. Project Reference

This dataset was created as part of the KASAL (Key-Axis-based Symmetry Axis Localization) Project.

You can find more details and access the project's GitHub repository here: https://github.com/WangYuLin-SEU/KASAL

5. License

This dataset consists of two parts with different licenses:

ShapeNetV2 data: The 3D models are under the original ShapeNetV2 license. Please refer to the ShapeNet page for the applicable license. Symmetry axis data: The symmetry annotations provided in this dataset are released under the MIT License.

6. Contributors

Yulin Wang (Southeast University, China)