CEAMLS Summer Research Program 2026

Learn Damage Assessment and FastAPI side by side.

Two complete learning tracks. The Damage Assessment track covers xBD dataset preprocessing, ResNet-50 + U-Net, ViT-B/16, Siamese change detection, and ONNX export. The FastAPI track teaches you to build the web API that serves those models — from your first endpoint to production deployment.

Research Track

Damage Assessment

Build a full satellite-image damage-classification pipeline. Start with 16-bit GeoTIFF preprocessing, then train and compare ResNet-50, ViT, and Siamese architectures on the xBD dataset.

API Track

FastAPI

Build a modern Python web API to serve damage-classification models. Learn routing, Pydantic schemas, file uploads, ONNX inference, and ATC-20 placard mapping.

Quick links

Jump straight to the explorer, quiz, curriculum, or your progress dashboard.

Core concepts, in plain English

FastAPI

A modern Python framework for building web APIs. The server FastAPI runs is what the browser talks to when you upload an image.

Endpoint

A URL the server knows how to answer — e.g. GET /health returns server status, POST /classify accepts an image and returns a damage class.

Pydantic schema

A class that declares the exact shape of request and response data, so invalid input is rejected with a clear error before reaching your code.

ONNX

A portable file format for trained models. Loading model.onnx is faster and requires no PyTorch on the server.

ATC-20 placard

An official post-earthquake/disaster tag system. Engineers post GREEN (safe), YELLOW (restricted), or RED (unsafe) on a building.

Siamese Network

A twin-branch neural net that learns to compare two inputs by pushing similar pairs together and dissimilar pairs apart in embedding space.

ViT

Vision Transformer — splits an image into patches and processes them with self-attention instead of convolution, excelling at global context.

FastAPI end-to-end pipeline

A quick visual of how the FastAPI app processes a satellite image from upload to ATC-20 placard.

Duration
5 weeks
Level
Beginner
Stack
FastAPI · ONNX · Rasterio
End-to-end pipeline
  1. Image Upload
    GeoTIFF or JPG/PNG
  2. File Validation
    Magic bytes + size
  3. Preprocessing
    Bands, normalize, tensor
  4. AI Inference
    ONNX Runtime forward
  5. %
    Damage Class
    Softmax + entropy
  6. ATC-20 Placard
    GREEN / YELLOW / RED
  7. Recommendations
    Engineer action