sunana/Huperflow

README

Overview

This project contains human perceived optical flow/motion data derived from images obtained from several third-party benchmark datasets. The images and labels used in this project originate from these publicly available datasets. Before using any images or labels, please consult and comply with the respective licenses and usage agreements provided on their official websites.

This project is intended for research and non-commercial use only. If you plan to use any of the data for commercial purposes, you must first obtain proper authorization from the respective dataset providers.

Third-Party Datasets and Reference Websites

The project makes use of the following 10 benchmark datasets:

1. KITTI

   • Website: http://www.cvlibs.net/datasets/kitti/
   • Description: The KITTI dataset, developed by the Karlsruhe Institute of Technology and its industry partners, is widely used for stereo vision, optical flow, and SLAM research.

2. Virtual KITTI 2

   • Website: https://europe.naverlabs.com/Research/Computer-Vision/Proxy-Virtual-Worlds/Virtual-KITTI-2
   • Description: Virtual KITTI 2 is a synthetic version of the KITTI dataset, offering controllable conditions and diverse scenarios for optical flow and scene understanding experiments.

3. MPI Sintel (NV version)

   • Website: https://sintel.is.tue.mpg.de/
   • Description: Based on the open-source animated film "Sintel," this dataset is designed for optical flow evaluation with challenging motion, occlusions, and realistic rendering.

4. VIPER

   • Website: https://playing-for-benchmarks.org/dataset_viper.html
   • Description: VIPER (Virtual Panoramic Environment Rendering) provides synthetic urban driving scenarios with comprehensive annotations, including optical flow, for multi-task vision research.

5. Spring

   • Website: https://lmb.informatik.uni-freiburg.de/resources/datasets/SceneFlowDatasets.en.html
   • Description: Part of the Scene Flow Datasets, Spring offers synthetic data generated from 3D scenes, which is useful for training and evaluating optical flow and related algorithms.

6. Monkaa

   • Website: https://lmb.informatik.uni-freiburg.de/resources/datasets/SceneFlowDatasets.en.html
   • Description: Also from the Scene Flow Datasets, Monkaa provides synthetic sequences from a Blender open-source project, suitable for optical flow, stereo matching, and depth estimation research.

7. MHOF

   • Website: https://humanflow.is.tue.mpg.de/
   • Description: The MHOF dataset is focused on multi-human optical flow. Users should consult the original resource for complete details regarding its usage and licensing.

8. Driving

   • Website: https://lmb.informatik.uni-freiburg.de/resources/datasets/SceneFlowDatasets.en.html
   • Description: The Driving dataset, also part of the Scene Flow Datasets, provides synthetic driving sequences with annotations for optical flow and other computer vision tasks.

9. FlyingThings3D

   • Website: https://lmb.informatik.uni-freiburg.de/resources/datasets/SceneFlowDatasets.en.html
   • Description: FlyingThings3D is a large-scale synthetic dataset featuring numerous 3D objects in motion. It is designed to offer challenging optical flow scenarios and support research in 3D scene understanding.

10. TartanAir

    • Website: https://theairlab.org/tartanair-dataset/
    • Description: Developed by the AIR Lab at Carnegie Mellon University, TartanAir offers a diverse range of indoor and outdoor synthetic environments with annotations for optical flow, depth, and semantic segmentation, suitable for multi-task learning.

Data Usage Guidelines

1. Third-Party Data Rights:

   • The images and labels provided in this project are subject to the licenses and usage agreements of their original sources.
   • Please review the official documentation or websites for each dataset to ensure compliance with their terms of use.
   • For commercial usage, secure the necessary permissions from the dataset providers.

2. Project-Specific Data:

   • Any additional annotations or modifications (e.g., the human perceived optical flow/motion labels) generated in this project are provided under the project’s own license (refer to the LICENSE file for details).
   • When citing or reusing this project’s contributions, please give proper attribution.

3. Disclaimer:

   • This project is not responsible for any legal or operational issues arising from the use of third-party data.
   • Users are solely responsible for ensuring that their use of the data complies with all relevant laws, guidelines, and licenses.

Reference & Contact

Please visit the provided links for the latest details on dataset usage policies, updates, and additional resources. Thank you for your interest in this project and for respecting the intellectual property rights of the original dataset providers.

Project Web: https://github.com/KUCognitiveInformaticsLab/Huperflow-Website/tree/main