Emotion Recognition in Context using different Transfer Learning Model

The ability to read and understand the emotions of a human is a matter of great importance in social interactions and further boosts the interaction between machines and humans. Despite tremendous progress in the field of artificial intelligence, there is no machine to fully comprehend the complications of feelings involved in a human being. This paper utilizes the EMOTIC dataset that comprises images of people in diverse natural conditions, and each of these images has annotations of the inferred emotion of the individual in the picture. This research focuses on an algorithm that should be able to automatically identify, extract, and recognize human emotions from images. It does this by analyzing facial expressions that convey a universal set of emotions shared among humans and incorporating contextual information from the scene into the data about the bounding box that identifies the person in the image. In this paper, we implement various CNN models for emotion recognition by focusing on the scene context’s role in emotion detection tasks. Some of the CNN models in the literature involve ResNet-50, VGG19 and DenseNet-169. This work describes the potential application of transfer learning on emotion recognition under the constraints of the dataset.