Enhancing Super-Resolution Models
Published · IPMV 2025This is a research paper I wrote comparing three GAN-based image super-resolution models: Real-ESRGAN, AESRGAN, and ESRGAN. It was published at IPMV 2025, the International Conference on Image Processing and Machine Vision, where I presented it in person. The paper is also available in the National High School Journal of Science and on ResearchGate, both linked above.
The problem
Super-resolution models that upscale low-resolution images each make different trade-offs between sharpness, realism, and artifacts, and it is not always clear which model is the right choice for a given case. I wanted to study those trade-offs directly.
My goal
To run a careful, side-by-side comparison of Real-ESRGAN, AESRGAN, and ESRGAN, and to describe where each model performs well and where it falls short.
Approach
The paper analyzes the three architectures and compares their super-resolution results, looking at how each handles detail and artifacts so the strengths and weaknesses of each approach are clear.
Publication and presentation
The work was published at IPMV 2025, and I presented it at the conference. You can read the full paper through the links above.