Document Type



KSU Electric Vehicle student competition Team (EVT)


Coneslayer Dataset


The Kennesaw State University's undergraduate Electric Vehicle (racing) Team, EVT, has created a lightweight neural network called "Coneslayer" for detecting orange traffic cones. As a very small model, Coneslayer has only around 6 million parameters, making it ideal for use on edge devices where processing power is limited.

Using a custom dataset of over 11,000 training images, we achieved good inference performance and accuracy for a diverse array of cone shapes and sizes while maintaining a low rate of false positives

To make our results reproducible, such as for training larger more accurate models, we are sharing this custom dataset.


The dataset is composed of two parts:

The first part is a homogeneous datset of about 10,000 images and labels obtained from track video screenshots and a subset of the FSOCO dataset

The second part is a heterogeneous dataset of about 600 images and labels from diverse real-world scenes obtained from dashcam video and smartphone pictures

About 2600 of the images from the homogeneous dataset were hand-labeled by members of the KSU EVT team. These were mixed at a 1 : 3 ratio with the pre-labeled subset of the FSOCO subset containing orange traffic cones.


Coneslayer Dataset

File Format


Creative Commons License

Creative Commons Attribution 4.0 International License
This work is licensed under a Creative Commons Attribution 4.0 International License.


Article Location