Computational Impressionism: High-Fidelity Art Synthesis via Transfer Learning in Resource-Constrained Environments
2026 10th International Artificial Intelligence and Data Processing Symposium (IDAP), Malatya, Türkiye, 5 - 06 Eylül 2026, ss.1-6, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/idap71869.2026.11728177
- Basıldığı Şehir: Malatya
- Basıldığı Ülke: Türkiye
- Sayfa Sayıları: ss.1-6
- Atatürk Üniversitesi Adresli: Evet
Özet
Artificial intelligence has continued to advance at a sustained rate in recent years. At this stage of development, generative adversarial networks (GANs) in particular have become a milestone in digital art production thanks to the realistic results they produce. Various studies exist in the field of art; however, studies demonstrating that it is possible to produce art despite limited data are few and unsatisfactory. The main goal in this study is to create high-resolution and aesthetically pleasing images. Achieving this goal requires significant hardware power. In this study, we presented an optimal framework called Computational Impressionism that produces original impressionist artworks despite limited data sets and limited hardware resources. In this study, we used the StyleGAN2-ADA architecture to solve the problems of limited data and training instability. We combined this architecture with a transfer learning approach, utilising weights trained on human faces dataset. Comparative experiments conducted at different resolutions revealed that transfer learning significantly contributes to the training process and enables faster learning of artistic touches such as brush strokes. The most important point of this study is that, in verifications performed on an NVIDIA A100 GPU, it achieved a very acceptable FID score of 57.66 at a resolution of 512x512, demonstrating that it is possible to create art without supercomputers. This result paves the way for the idea that producing much more original artworks with artificial intelligence will be possible in the future. Furthermore, it will also bring to mind the notion that generative artificial intelligence can make significant contributions to art.