Neural Network Training: Watermarks Removal
APS360 Artificial Intelligence Fundamentals
-
Implemented a random watermark generator that could add watermarks to images with random font, size, location, and shape
-
Designed the architecture of the convolutional autoencoder and performed deep neural network training
-
Achieved a final testing loss of 0.0112 between the original and processed image sets
-
Provide a guideline for organizations to develop watermarks that better protect their copyrights
-
Led a team of four to compose a professional report and a video presentation
Overall Model of the Watermark Removal Project
Project Details
Structure of the watermark generator and sample results
Architecture of the primary model (convolutional autoencoder), the baseline model (encoder), and the training curve
Visualization of reconstructed image compared to the watermarked version
