Automatic Bug Inference via Deep Image Understanding

Abstract

In mobile crowdsourced testing, crowdworkers are usually far from experts, and low-quality bug reports are submitted, in which the bug descriptions are usually poorly written. Thus, the bug descriptions are hard to read and helpless for bug inference and bug understanding. To ease the understanding of bug scenarios, we present a novel method called BIU (Bug Inference via Image Understanding), which employs image understanding techniques to help crowdworkers automatically infer bugs and generate bug descriptions using bug screenshots. In this way, the burden of crowdworkers will be lowered and their working efficiency and report quality will be greatly improved. According to our preliminary experiments, the accuracy of BIU can reach up to 90%. The demonstration video can be found at: https://youtu.be/ZBOIqtdRFaU.