Extreme Image Compression using Fine-tuned VQGAN Models