Library Digital Collections

Data from: Multi-Source Feature Fusion for Object Detection Association in Connected Vehicle Environments

View Collection Items

Collections »

Data from: Multi-Source Feature Fusion for Object Detection Association in Connected Vehicle Environments

About this collection

Extent

1 digital object.

Cite This Work

Thornton, Samuel; Flowers, Bryse; Dey, Sujit (2023). Data from: Multi-Source Feature Fusion for Object Detection Association in Connected Vehicle Environments. UC San Diego Library Digital Collections. https://doi.org/10.6075/J0HX1CVJ

Description

This dataset contains virtual and real-world examples of object detection association in connected vehicle environments. The main object class that is represented in this dataset is vehicles and these represent the majority of samples within the dataset, though some other classes such as motorcycles, bicycles, trucks, and buses are also present in the dataset. The image data are RoI images produced by a machine learning object detector; every pair of images for each timestep is assigned a positive label ("1") if both RoI images are of the same vehicle, otherwise a negative association label ("0") is assigned to the image pair.

There are 10k+ real-world image pairs that have been human labeled as well as over 5 million virtual images pairs that have been labeled by a heuristic automatic labeler. The real-world data was recorded on the UCSD campus using Intel RealSense D415 stereo cameras and the Sighthound vehicle recognition API was used to detect and classify each object. The virtual data was recorded in the Carla Simulator (https://carla.org/) and the object detector used was detectron2. Position estimates are included for the virtual data using information from Carla as well as the object detector (see paper for more details).

Date Collected
  • 2020 to 2022
Date Issued
  • 2023
Authors
Principal Investigator
Technical Details

The data consists of .png files for the images and .npy files for the data labels. For the data generation and Data_Loader.ipynb, we used python 3.9, pytorch 1.10, and numpy 1.21 but any of the same major versions should work as well (i.e. python 3.x, pytorch 1.x, numpy 1.x). Opencv and matplotlib are also used for plotting and viewing images.

Funding

This work was funded by the UCSD Center for Wireless Communications (CWC)

Geographic
Topics

Formats

View formats within this collection

Identifier

Identifier: Bryse Flowers: https://orcid.org/0000-0003-1474-4732

Identifier: Samuel Thornton: https://orcid.org/0000-0002-3087-2577

Identifier: Sujit Dey: https://orcid.org/0000-0001-9671-3950

Related Resources