DeepFake Image Detection and Classification using EfficientNet Model

Deep learning models such as Vision Transformer (VIT) and EfficientNet, have brought significant advancements to computer vision tasks like image classification, object detection, and image generation. In this paper, comparative analysis of VIT and EfficientNet models has been done with more attention of on their architectural disparities, training procedures, and performance characteristics. Deep learning models like Vision Transformer (VIT) and EfficientNet have revolutionized computer vision. VIT uses self-attention techniques instead of convolutional layers to capture global relationships but with higher computational. EfficientNet models, with compound scaling offer a trade-off between accuracy and efficiency. EfficientNet models are computationally fast with competitive accuracy and is suitable for resource-limited contexts. The paper suggests choosing the best model based on specific use cases and resource limitations. Quantitative analysis of the present work has been done using confusion matrix. It has been observed that EfficientNet models are providing higher performance ratio.