Sequence-Based Image Prediction using GAN and LSTM

Video frame prediction is a fundamental task in computer vision with applications in autonomous systems, surveillance, and video compression. Traditional methods, such as convolutional LSTMs and Variational Autoencoders (VAEs), often suffer from temporal inconsistencies and blurry outputs This paper proposes a GAN-based framework with convolutional layers for generating sharp and temporally coherent video frame. The model consists of a CNN-based generator and discriminator, trained adversarially to enhance prediction accuracy. Optimized with mixed precision training and the Adam optimizer, the model achieves 99.14 % accuracy, outperforming existing approaches. A comparative analysis demonstrates significant improvements in clarity, efficiency, and computational cost. While the model performs well on the Moving MNIST dataset, future work will focus on extending it to real-world datasets and integrating attention mechanisms for improved motion prediction.