U-Net deep learning colourisation of greyscale images¶
"U-Net deep learning colourisation of greyscale images" explores the application of deep learning techniques to transform grayscale images into colorized versions. Utilizing a U-Net architecture with a ResNet-34 encoder pretrained on ImageNet, the model employs a feature loss function based on VGG-16 activations, pixel loss, and gram matrix loss to achieve high-quality colorization. The Div2k dataset serves as the training and validation source, with data augmentation techniques such as random cropping, horizontal flipping, lighting adjustments, and perspective warping enhancing the model's robustness.
The article demonstrates how time, resources and money can be saved with fine tuning existing models.
The training process involves converting original RGB images to grayscale for input, with the original RGB serving as the target output. By leveraging a pretrained encoder, the model benefits from existing feature detection capabilities, significantly accelerating the training process. The feature loss function, incorporating VGG-16 activations, enables the model to capture intricate details and textures, resulting in more realistic colorization. This approach contrasts with Generative Adversarial Networks (GANs), offering faster convergence to optimal results.
The article showcases several examples demonstrating the model's effectiveness in colorizing grayscale images, highlighting its potential applications in image restoration and enhancement. The combination of U-Net architecture, ResNet-34 encoder, and a sophisticated loss function contributes to the model's ability to produce visually compelling colorized images. The author also references related works, such as Jason Antic's DeOldify model, indicating a broader interest and ongoing development in the field of image colorization using deep learning techniques.
Read the full article here: