The model ranks photos aesthetically and assigns a score for each image on a scale of 1 to 10. The scoring is directly in line with how the training data is typically captured, making it a predictor of human preferences.
such as increase in brightness, highlights, and shadows to increase chances of getting a higher score
Provide feedback to jobseekers
based on the client’s suggested norms of uploading profile pictures, such as remove sunglasses.
While the client has excellent data-driven insights that help families narrow their candidates based on factors such as pay rates, location, family makeup and needs, the photo review process for the profile pictures for all caregivers’ accounts was manual. The customer service executives would manually skim through all the profile pictures uploaded to accept or reject them.
We sought to automate this process of scoring and tagging all the caregivers’ profile pictures, based on some of the universally accepted aesthetic norms by introducing a deep CNN model.
Aesthetic quality assessment of photos is a challenging task, especially when taking into account the complexity of various photos and the subjectivity of human’s aesthetic perception. To solve this, we introduced a deep CNN model which can encode information regarding standard image quality (technically), and facial attributes (aesthetically), and continue to train it to combine these two to explicitly predict the aesthetics of the profile pictures.