Fuzzy Windows with Gaussian Processed Labels for Ordinal Image Scoring Tasks

In this paper, we propose a Fuzzy Window with the Gaussian Processed Label (FW-GPL) method to mitigate overlap problem in neighboring ordinal category when scoring images. Many published conventional methods treat challenge as traditional regression and make strong assumption that each owns an adequate intrinsic rank outline its distribution. Our FW-GPL aims refine label pattern by using two novel techniques: (1) assembling fuzzy logic fully connected layer of convolution neural networks (2) transforming labels process. Specifically, it incorporates heuristic from characteristic simultaneously plugs distribution shapes penalize difference between targeted neighbors ensure concentrated regional Accordingly, function these proposed windows is leveraged minimize influence majority classes mislead prediction minority samples. model specifically designed carefully avoid partially missing continuous facial-age segments. It can perform competitively whole dataset. Extensive experimental results on three facial-aging datasets one ambiguous medical dataset demonstrate our achieve compelling performance compared State-Of-The-Art (SOTA).