Visual Glitches Classification for Video Game Using Deep Learning-Based Techniques

Glitches, unintended and anomalous behaviors in video games, often degrade player experiences. Our objective is to develop a glitch classification model that can accurately identify and categorize different types of visual glitches. The proposed model incorporates Neural Architecture Search (NAS) to identify the optimal architecture and hyperparameters for the classification of these anomalies within in-game visual. We construct a unique, comprehensive dataset composed of numerous instances of common visual glitches along with non-glitched game footage. The proposed deep learning model is then trained and evaluated on this dataset, and our results demonstrate a notable enhancement in glitch detection accuracy 88% compared to conventional. Experimental results demonstrate the potential of the visual glitches classification model.