Research on Computer Vision Teaching Reform Based on Productive Failure Cases

Abstract

In response to issues such as the limited algorithm debugging proficiency and the disjunction between theory and practice in computer vision courses, this paper puts forward a teaching reform model grounded in the "Productive Failure" theory. Through the construction of a three - tier hierarchical case library encompassing code - level errors, algorithm logic deficiencies, and model tuning challenges, a reverse teaching path of "Failure Demonstration - Root Cause Analysis - Solution Iteration" is designed. This path is then integrated with a dynamic evaluation mechanism to monitor students' competence development.The case library consolidates high - frequency errors from student experiments. It combines open - source tools and online platforms to realize dynamic updates and hierarchical adaptation of cases. Practical implementation indicates that this model notably enhances students' engineering practice capabilities. Specifically, 84% of students can independently rectify code - level errors, with the average debugging efficiency increasing by 28%. In model tuning tasks, 32% of students proposed innovative solutions, among which 19 items were incorporated into the iterative case library.This reform validates the efficacy of "Controlled Failure" in cultivating technology transfer and innovation abilities, offering a replicable model for the practical teaching of computer vision courses.