the analysis code for CS236 project
The machine learning training and inference were conducted on the Web UI platform: https://github.com/AUTOMATIC1111/stable-diffusion-webui The LoRA model and the report for the pixel-art style is in this shared Google Drive: https://drive.google.com/drive/folders/16BTs-uk4XDWfeYNcp65Tfj4kDn654zTf?usp=sharing
frequency_analysis: compute the intensity of the high-frequency signals in the image using FFT. The input is the image name of the analysis target.
compute_gradient: use the Sobel filter to detect the edge for the given image.
plot_gradient: process, analyze, and visualize the edges detected in the image.
It processes the LoRA training results in the safetensor file format and computes the subspace similarity between the selected targets.
It converts a 512x512 image generated by the diffusion model into a 64x64 image usable as the game asset.