Mask-Benchmark Dataset

Project Page | Paper | Code

This repository contains the dynamic scene novel-view segmentation benchmarks used in the paper "TRASE: Tracking-free 4D Segmentation and Editing" (also referred to as "SADG: Segment Any Dynamic Gaussian Without Object Trackers"). The benchmarks are designed for evaluating segmentation performance in dynamic novel view synthesis across various datasets.

Overview

The Mask-Benchmark dataset provides ground truth segmentation masks for multiple dynamic scene datasets, including:

• HyperNeRF (A Higher-Dimensional Representation for Topologically Varying Neural Radiance Fields, ACM Transactions on Graphics (TOG))
• NeRF-DS (NeRF-DS: Neural Radiance Fields for Dynamic Specular Objects, CVPR 2023)
• Neu3D (Neural 3D Video Synthesis from Multi-view Video, CVPR 2022)
• Google Immersive (Immersive Light Field Video with a Layered Mesh Representation, SIGGRAPH 2020 Technical Paper)
• Technicolor Light Field (Dataset and Pipeline for Multi-View Light-Field Video, CVPRW 2017)

These benchmarks allow for quantitative evaluation of segmentation accuracy (mIoU and mAcc) in novel view synthesis for dynamic scenes, which was previously lacking in the field.

License Information for Mask-Benchmark Dataset

This Mask-Benchmark dataset is primarily licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC-BY-NC 4.0).

You are free to:

• Share — copy and redistribute the material in any medium or format
• Adapt — remix, transform, and build upon the material

Under the following terms:

• Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made
• NonCommercial — You may not use the material for commercial purposes

For the full license text, please visit: https://creativecommons.org/licenses/by-nc/4.0/legalcode

Component Datasets and Their License Terms

The Mask-Benchmark incorporates data derived from multiple source datasets, each with their own license terms that must be respected:

1. Neural 3D Video Dataset (Neu3D)

Licensed under CC-BY-NC 4.0.

2. HyperNeRF Dataset

Licensed under Apache License 2.0.

3. NeRF-DS Dataset

Licensed under Apache License 2.0.

4. Google Immersive Dataset

Refer to the original license terms provided by the Google Immersive project.

5. InterDigital Light-Field Dataset (Technicolor)

INTERDIGITAL LIGHT-FIELD DATASET RELEASE AGREEMENT

The goal of the InterDigital Light-Field dataset is to contribute to the development and assessment of new techniques, technology, and algorithms for Light-Field video processing. InterDigital has copyright and all rights of authorship on the dataset and is the principal distributor of the Light-Field dataset.

CONSENT The researcher(s) agrees to restrictions including:

1. Redistribution: Shall not be further distributed without prior written approval.
2. Modification and Non Commercial Use: May not be modified or used for commercial purposes.
3. Publication Requirements: Permits publication for scientific purposes only.
4. Citation/Reference: All documents must acknowledge use by citing: Dataset and Pipeline for Multi-View Light-Field Video. N. Sabater, et al. CVPR Workshops, 2017.

Using the Mask-Benchmark Dataset

By using the Mask-Benchmark dataset, you agree to:

1. Comply with the CC-BY-NC 4.0 license governing the overall dataset.
2. Adhere to all component dataset license terms listed above.
3. Properly cite both the Mask-Benchmark and the original source datasets.
4. Use the dataset for scientific and research purposes only.

How to Use Mask-Benchmark Dataset

Please follow the step in our code to download and unzip Mask-Benchmark.zip. Please note that for evaluation, only Mask-Benchmark.zip is used, the other subfolders are only for HF dataset viewer for visualization purpose.

BibTex

@article{li2024trase,
    title={TRASE: Tracking-free 4D Segmentation and Editing},
    author={Li, Yun-Jin and Gladkova, Mariia and Xia, Yan and Cremers, Daniel},
    journal={arXiv preprint arXiv:2411.19290},
    year={2024}
}