Yonsei Stress Image Database

YonseiStressImageDatabase is a database built for image-based stress recognition research. We designed an experimental scenario consisting of steps that cause or do not cause stress; Native Language Script Reading, Native Language Interview, Non-native Language Script Reading, Non-native Language Interview. And during the experiment, the subjects were photographed with Kinect v2. We cannot disclose the original image due to privacy issues, so we release feature maps obtained by passing through the network. We used ResNet-18 as the network for extracting the feature maps and released feature maps after the 3rd block out of a total of 4 blocks. The input to the network is a 112×112×3 face RGB image and the size of the feature maps is 7×7×256 (height×width×channel). This database consists of 50 subjects and 2,020,556 data. There are two tasks that can be done with this data. First, there is 4 class classification that classifies the data of each experimental stage, and there is 3 class classification that put the data of the two 'Script Reading' stages as one class.