Real-Time Performance Optimization in Digital Art Animation Stylization Using Self-Supervised Learning

To address the problems of low computational efficiency and poor real-time performance in the stylization process of digital art animation, a self-supervised learning algorithm is introduced to optimize the performance of the stylization process and improve real-time rendering efficiency. First, a self-supervised learning framework is designed to automatically learn stylized features through a generative model to reduce the reliance on manual labels. Then, a contrastive learning method is used to optimize the model's ability to extract style features, making the stylized results more accurate. Finally, by integrating real-time rendering technology, the real-time performance of stylization processing is further improved, and smooth conversion of dynamic images is achieved. In the impressionist style, the SSIM (Structural Similarity Index Measure) value using contrastive learning is 0.92, which is significantly higher than 0.85 without contrastive learning. The PSNR (Peak Signal-to-Noise Ratio) also increases from 32.1 dB to 35.6 dB, indicating that contrastive learning significantly improves the structural similarity and detail retention capabilities of images. The application of self-supervised learning in digital art animation stylization not only solves the contradiction between efficiency and accuracy, but also provides a new technical route for real-time animation rendering, which has great application prospects.