AI-Assisted Task-Driven Reform of a Computer Vision Course: A Hierarchical Practice Framework for Programming Competence
TL;DR
This paper proposes an AI-assisted, task-driven reform framework for a computer vision course that provides a practical model for integrating human-AI collaboration into specialized AI courses while preserving students' independent reasoning, programming competence, and experimental accountability.
Abstract
The rapid development of artificial intelligence (AI) is reshaping not only educational content but also the organization of classroom practice, learning support, and assessment. Computer vision courses are typical of this transformation: they require students to understand complex models while also completing data processing, programming, training, debugging, and evaluation tasks. In response to common problems such as fragmented experiments, insufficient programming support, and unregulated use of generative AI tools, this paper proposes an AI-assisted, task-driven reform framework for a computer vision course. The framework uses virtual try-on with traditional ethnic costume imagery as a culturally situated mainline task and organizes practice into three progressive levels: code reproduction, model modification, and independent design. It also introduces traceable AI use, structured experimental reporting, and multi-source assessment. The reform provides a practical model for integrating human-AI collaboration into specialized AI courses while preserving students' independent reasoning, programming competence, and experimental accountability.