BUILDING AND OPTIMIZING LIGHTWEIGHT GENERATIVE ADVERSARIAL NEURAL NETWORKS TO ENHANCE VIDEO QUALITY IN THE CLIENT DEVICES USING WEBGPU
M. R. Maksymiv, Тарас Рак · Komp'ûternì sistemi ta merežì · 2025
The paper considers problems for the tasks of improving the quality of digital video images for cloud environments, as well as on the client side using generative adversarial neural networks (GANs) adapted for work in the browser. A method is proposed that uses WebGPU for accelerated execution of convolutional calculations, which allows to increase the resolution and improve the quality of low-quality video in real time without significant load on servers. Optimization of the neural network includes the use of Pruning and Knowledge Distillation methods, which made it possible to reduce the size of the model by 40–60% without significant loss of quality. The results of the experiments showed that the implementation of the proposed method increases the performance of video processing in the browser by 2–4 times compared to models based on the WebGL interface. The video quality assessment showed an improvement in PSNR and an increase in SSIM compared to traditional methods of increasing resolution. The proposed approach can be integrated into streaming services and web applications, which will reduce the load on computer networks and provide a better user experience with lower costs for cloud and server computing. Key words: Generative adversarial networks, High definition video, High-resolution imaging , Image Superresolution, Optimization models, Neural network