[UNI]101: An Educational Dataset for Introductory Computer Vision

Ethan Seefried, Changsoo Jung, Videep Venkatesha, Trevor Chartier, Caleb Christian, Jack Fitzgerald, Mariah Bradford, Sifatul Anindho, Matthew Sturgeon, Nathaniel Blanchard

Keywords: Computer Vision, Education, In the Wild, Object Detection, Classification, Gaussian Splatting

TL;DR: A dataset designed for university classroom computer vision education.

Abstract: Education in Machine Learning, particularly in Computer Vision, currently lacks modern, relatable datasets specifically designed to support a structured learning progression. Existing datasets such as MNIST or ImageNet are cornerstones in teaching Computer Vision, but they address isolated tasks and may not resonate with students' daily experiences. Addressing this gap, we introduce [UNI]101, a comprehensive video dataset consisting of 277,056 frames of university buildings, captured at 24 FPS in 1080p resolution. [UNI]101 is designed to scaffold an entire semester of introductory Computer Vision: students begin with image classification across 48 campus buildings, advance to object detection over five commonly seen campus object classes (bike racks, trash cans, doors, electrical boxes, and light posts), and progress to 3D scene reconstruction via Gaussian Splatting, all using a single, familiar dataset. The dataset includes 750 unique labeled objects, benchmark models for each task (six for classification, five for detection), and educational Jupyter Notebooks structured as lesson plans. By grounding an entire curriculum in scenes students encounter daily, [UNI]101 bridges the gap between foundational Computer Vision education and modern research topics, providing a progressive, hands-on learning experience within a unified framework.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Track: Proceeding Track

Submission Number: 11