Method for Out-of-Distribution Data Detection for AR Teaching Based on Semi-supervised Deep Learning Network

Out-of-distribution detection is an important task in image processing and plays an important role in AR Teaching image recognition. However, when the out-of-distribution data input into deep learning networks, the results tend to have a uniform distribution across various label categories, so it is difficult to detect out-of-distribution data using DNN. We propose a method that the intermediate layer outputs of neural networks are used as features for re-input into the neural network for the recognition of out-of-distribution data. Experimental results demonstrate that by training with a small amount of out-of-distribution data, the proposed method improves the AUROC metric by 9%.