MATLAB-BASED APPROACH TO INVESTIGATING DATASET TESTING AND TRAINING FOR ENHANCED HUMAN-LIKE FRAME PREDICTION USING CONVOLUTIONAL NEURAL NETWORKS IN DIVERSE SCENES THROUGH DEEP LEARNING

Deep Learning has been applied to train the convolutional neural networks (CNNs) for accurate frame prediction. Using CNNs, a MATLAB-Based approach is used to investigate dataset testing and training techniques for obtaining enhanced human-like frame prediction using CNNs in diverse scenes. Furthermore, studies were explored in Deep Learning, which is a kind of Machine Learning that can be trained, supervised, semi-supervised and unsupervised. Specifically, the proposed study applies deep learning methods, including Convolutional Neural Networks (CNNs), for next-frame prediction. The Catz Dataset is utilized as the training data for this investigation. The experimental results reveal that CNNs can indeed be used to achieve human-like frame prediction in diverse scenes. The best performing model, a hybrid CNN and LSTM network, exhibits a significantly improved perceptual distance rating of 26.7127, outperforming the initial CNN model. These findings demonstrate the potential of CNNs trained using deep learning techniques for accurate frame prediction tasks. The study has also shown that impact of the training and testing ratios on the performance of an enhanced human-like frame prediction using CNNs and MATLAB. The experiments through MATLAB have shown that higher training percentage means that a larger portion of dataset for training the model have been used while a lower training percentage shows that a large portion of the dataset reserved for testing the model's performance.