Related Experiment Video
Updated: Jul 4, 2025
03:31
Author Spotlight: Enhancement of Salient Object Detection for Smart Grid Applications
Published on: December 15, 2023
542
A neuromorphic dataset for tabletop object segmentation in indoor cluttered environment.
Xiaoqian Huang1,2, Sanket Kachole3, Abdulla Ayyad1
1Advanced Research and Innovation Center (ARIC), Khalifa University, Abu Dhabi, UAE.
Scientific Data
|January 25, 2024
Summary
Researchers introduce the Event-based Segmentation Dataset (ESD), a novel 3D dataset for benchmarking event-based segmentation algorithms. ESD provides crucial depth and instance labels for cluttered indoor scenes, advancing computer vision research.
Area of Science:
- Computer Vision
- Robotics
- Sensor Fusion
Background:
- Conventional cameras struggle with motion blur, low dynamic range, and temporal sampling limitations.
- Existing event-based datasets lack comprehensive depth information crucial for occluded scene segmentation.
Purpose of the Study:
- Introduce a novel, high-quality event-based dataset for benchmarking segmentation algorithms.
- Provide a 3D spatial-temporal dataset with critical depth and instance labels for indoor object segmentation.
Main Methods:
- Developed the Event-based Segmentation Dataset (ESD) using stereo-configured event-based cameras.
- Collected 145 sequences with 14,166 annotated RGB frames and millions of events.
- Annotated event-wise depth and instance labels for tabletop objects.
Main Results:
- ESD offers a unique 3D spatial-temporal benchmark for event-based segmentation.
- The dataset includes dense annotations for RGB, depth, and event data.
- Presents a challenging benchmark for segmenting objects in cluttered indoor environments.
Conclusions:
- The release of ESD aims to foster advancements in event-based segmentation research.
- Provides a valuable resource for developing and evaluating algorithms requiring depth information.
- Encourages the development of more robust segmentation techniques for real-world applications.
