DiT (Diffusion Transformer) Project — Quick Guide

NOTE: report.pdf contains the comprehensive report.

This repository contains scripts for training, analyzing, and visualizing Diffusion Transformer (DiT) models for image generation.

Script Overview

dit_models.py: Core DiT model classes and utilities. Import this in other scripts; do not run directly.
run_experiments.py: Trains DiT models from scratch with different hyperparameters. Saves results and checkpoints.
- Run: python run_experiments.py
analyze_experiments.py: Analyzes experiment results and generates plots from the CSV output of training.
- Run: python analyze_experiments.py experiments/<timestamp>/experiment_results.csv
dit_pretrained_using_pixelart.py: Uses a pretrained PixArt-XL-2-512x512 model with DDIM sampling and classifier-free guidance (CFG). Visualizes the denoising process and CFG effects.
- Run: python dit_pretrained_using_pixelart.py
visualize_trained_models.py: Loads your trained DiT models and creates 10x10 grids showing the denoising evolution.
- Run: python visualize_trained_models.py
dit_pretrained_and_cfg_using_sd.py: Variant using Stable Diffusion as the backend for pretrained DiT-based visualization and CFG analysis.
- Run: python dit_pretrained_and_cfg_using_sd.py

Name		Name	Last commit message	Last commit date
Latest commit History 13 Commits
__pycache__		__pycache__
experiments/20251021_053158		experiments/20251021_053158
pixelart_results		pixelart_results
sd1.4_results		sd1.4_results
visualizations/trained_models		visualizations/trained_models
.gitignore		.gitignore
README.md		README.md
analyze_experiments.py		analyze_experiments.py
dit_cfg_using_pixelart.py		dit_cfg_using_pixelart.py
dit_models.py		dit_models.py
dit_pretrained_and_cfg_using_sd.py		dit_pretrained_and_cfg_using_sd.py
fid_evaluator.py		fid_evaluator.py
report.pdf		report.pdf
run_experiments.py		run_experiments.py
visualize_trained_models.py		visualize_trained_models.py