Learning to focus on region-of-interests for pain intensity estimation

The breakthrough success of many deep learning approaches is mainly due to the availability of large-scale labeled datasets. However, large-scale labeled datasets are not always available in some domains. Pain intensity estimation is unsurprisingly one those domains that suffer from lacking of labeled training data. In this work, we proposed a learning approach that is able to learn to focus on region-of-interests in face image for better feature extraction, thus improving overall performance of the network when training on a limited amount of data. Our extensive experiments demonstrate that our learning to focus on region-of-interests approach performs better in overall compared to state-of-the-art approaches in pain intensity estimation. From the experimental results, we emphasise the importance of learning to focus on region-of-interests for better extracting feature representations and reducing the effect of overfitting when training on a limited amount of data.