Dalton Laboratory

I joined the Dalton Laboratory in May 2025 and spent the summer working with PhD candidate Qinyi Tian on projects in the areas of super-resolution artificial intelligence models and porous material science. I learned to use new software like Dragonfly and ImageJ to visualize and edit micro-CT scans in 3D space and conducted comparative data analysis of alternative ML models for use in publication. The laboratory is directed by Laura Dalton, Assistant Professor in the Department of Civil and Environmental Engineering.

Publication Under Review: Image Super-Resolution Model

Qinyi Tian, Spence Cox, & Laura Dalton. CATformer: Contrastive Adversarial Transformer for Image Super-Resolution. Under review at AAAI Conference on Artificial Intelligence (AAAI-26).

For this project, I independently conducted inference using five established super-resolution models (ESRGAN, SwinIR, Real-ESRGAN, HAT, and DiffBIR) across four benchmark datasets (CelebA-HQ, Div2k, Urban100, and RealSR). I designed and executed a robust experimental pipeline to upscale low-resolution images, compute quantitative performance metrics (PSNR, SSIM, LPIPS, and MSE), and compare baseline models with the proposed CATformer model. I delivered these validated results to the PI for integration into the publication’s evaluation section.

Current Work: Stress-Strain Informed Neural Network

We aim to develop a stress-strain informed neural network to predict morphological indicators of porous materials. To train the model, I am 3D-printing and preparing porous specimens and conducting in-situ CT scans under uniaxial compression. I use ImageJ and Dragonfly to preprocess CT scans and the PyTorch library for model development.

Google Sites

Report abuse