TY - JOUR AU - Yan, H. AU - Achkar, A. AU - Mishra, Akshaya AU - Naik, K. PY - 2016/10/03 Y2 - 2024/03/29 TI - Automated Failure Detection in Computer Vision Systems JF - Journal of Computational Vision and Imaging Systems JA - J. Comp. Vis. Imag. Sys. VL - 2 IS - 1 SE - Articles DO - 10.15353/vsnl.v2i1.84 UR - https://openjournals.uwaterloo.ca/index.php/vsl/article/view/84 SP - AB - <p>Human validation of computer vision systems increase their operating<br />costs and limits their scale. Automated failure detection can<br />mitigate these constraints and is thus of great importance to the<br />computer vision industry. Here, we apply a deep neural network<br />to detect computer vision failures on vehicle detection tasks. The<br />proposed model is a convolution neural network that estimates the<br />output quality of a vehicle detector. We train the network to learn<br />to estimate a pixel-level F1 score between the vehicle detector and<br />human annotated data. The model generalizes well to testing data,<br />providing a mechanism for identifying detection failures.</p> ER -