This project, titled I’m Something of a Painter Myself, utilizes CycleGAN, a type of generative adversarial network (GAN), to replicate the artistic style of Claude Monet. The objective is to see if the model can generate images that are indistinguishable from Monet’s work to both viewers and image classifiers. Key steps and challenges in the project include:

• Dataset Preparation and Exploration: Images, primarily in TFRecord format, are preprocessed with functions that decode, normalize, and resize them. Augmentation techniques, such as rotation, flipping, and cropping, further diversify the training set, enhancing the GAN's ability to generalize.
• Model Architecture: The project involves building a CycleGAN, with two models—a generator and a discriminator. The generator is trained to create Monet-like images from input photos, while the discriminator distinguishes real Monet paintings from generated images. This adversarial setup drives the generator to improve continuously.
• Loss Functions and Tuning: The project defines various loss functions, including adversarial and cycle consistency losses, and explores parameter tuning to identify the optimal settings for learning rate, batch size, and epochs. Due to computational limits, the tuning focused on small batch sizes and lower epoch numbers.
• Evaluation: The model’s performance was primarily evaluated through visual inspection of the generated images and by monitoring the loss function values during training. A loss score of approximately 49.82 was achieved, which the user deemed satisfactory for a beginner's attempt. Although Fréchet Inception Distance (FID) is typically used for evaluating GANs, it was not employed in this initial version. Future iterations may incorporate FID to provide a more robust metric for assessing visual authenticity.