A Fine-Tuned MobileNetV3 Model for Real and Fake Image Classification
Gurpreet Singh, Kalpna Guleria, Shagun Sharma · 2024
This study focuses on the utilization of the MobileNet model for the classification of artificial and real images, addressing the increasing demand for accurate image identification in the digital realm. Through the implementation of convolutional neural networks (CNNs), specifically optimized for mobile and embedded devices, we aim to distinguish between authentic and digitally manipulated images efficiently. After training the MobileNet model on a dataset comprising both real and artificial images, we evaluated its performance on a validation set. The results indicate a validation accuracy of 95.02%, demonstrating the model's robustness in distinguishing between the two image categories. Additionally, the validation loss was found to be 0.1621, indicating minimal discrepancies between predicted and actual labels. Furthermore, the precision and recall metrics provide insights into the model's ability to identify real and artificial images accurately. With a validation precision of 0.9494 and a validation recall of 0.9511, the MobileNet model showcases high precision in correctly classifying both image types while minimizing false positives and negatives. Overall, these findings highlight the effectiveness of the MobileNet model in artificial image detection, offering a reliable solution for combating the proliferation of manipulated visuals in various domains, from social media to journalism and beyond.