ImageTRT

Image Treatment Effect Estimation using synthetic image generation and deep learning.
This project investigates how to estimate direct and indirect treatment effects when treatments are images, while accounting for spatial spillover effects and aggregation bias.

---

📂 Project Structure

• Synthetic Data Generation/
  Generate synthetic datasets for treatments and outcomes (e.g., wetlands, DEM, capital, outcome, ITE, theta).

• Model/

  • main.ipynb: Core notebook experimenting with different model architectures.
  • Integrates utilities for embeddings, regressions, and evaluation.

• Utils/
  Utilities for:

  • Image preprocessing (load_and_resize.py, convolution.py)
  • Synthetic treatment/outcome generation
  • Treatment & outcome regressions
  • Wetland selection
  • Evaluation functions for treatment effects

---

🚀 Approach

We test combinations of:

• Autoencoders
• Transformer Architectures
• CNN-based models
• Other embedding approaches

The objective is to determine the most effective way to capture direct and indirect causal effects in high-dimensional spatial data.

---

✅ Current Progress

• Early model experiments in Model/main.ipynb (CNNs, autoencoders, embeddings)
• Work-in-progress: spatial spillovers + aggregation bias integration