This project received an Excellent grade for its attempt to utilize different ML models and clear presentation.
-
Presentation.pdf: Presentation slides
-
Report.pdf: Final project report
-
Coding.twbx: Coding file
Image Authenticity Verification
In this project, we attempt to distinguish between real dog images and AI-generated dog images using a CNN model and a CLIP model. Both models are trained using a supervised machine learning approach. We then compare the performance of both models to see how effectively these supervised models perform in classifying the images.
The results show that CLIP comprehensively outperforms the traditional CNN across all evaluation metrics. Most notably, in detecting the rarer "AI Dog" category, CLIP delivers a major leap in stability and detection capabilities that traditional convolutional networks cannot achieve.
Python(Sckit-learn,Numpy,Tqdm,Matlibplot,Seaborn)
 |
 |
 |
 |
 |
 |
For more comprehensive analysis and detailed visualizations, please refer to the report or the coding file.