CS230 Project: Image-tampering Classification for Fake News Detection

Introduction

Growing research body has shown that image-tampering has become a salient strategy for fake news. To address this problem, we researched on ResNet-50 and VGG-16 models, implemented with the pretrained weights for ImageNet, and retrained on the CASIA dataset.

This repository stored the codes and results of this project.

Folder Structure

• Baseline: codes in using the baseline model (shallow neural network with three layers of CNNs)
• ResNet50: codes in using the ResNet-50 structure for transfer learning (using pretrained weights vs retraining, different hyperparameters)
• VGG16: codes in using the VGG-16 structure for transfer learning (using pretrained weights vs retraining, different hyperparameters)
• Results

Authors

• Veronica Wang
• Yingdan Lu

Acknowledgements

CS230 Teaching Team, Stanford University