Deep Regression Neural Networks for Proportion Judgment

Deep regression models are widely employed to solve computer vision tasks, such as human age or pose estimation, crowd counting, object detection, etc. Another possible area of application, which our knowledge has not been systematically explored so far, is proportion judgment. As a prerequisite for successful decision making, individuals often have use judgment strategies, with they estimate the magnitude one stimulus relative another (larger) stimulus. This makes this estimation problem interesting application machine learning techniques. In regard this, we proposed various deep architectures, tested on three original datasets very different origin and composition. novel approach, assumption that model can learn concept without explicitly counting individual objects. With comprehensive experiments, demonstrated effectiveness predict proportions real-life more reliably than experts, considering coefficient determination (>0.95) amount errors (MAE < 2, RMSE 3). If there no significant number in determining ground truth, an appropriate size dataset, additional reduction MAE 0.14 be achieved. The used will publicly available serve reference data sources similar projects.