README.md

SpaceNet 9 Baseline

Install

While in this directory, run

docker build -t spacenet9:1.0 .

With the image already built, start a container. For example:

docker run -it --gpus all --ipc host -v ./:/tmp/spacenet_baseline --rm --name my_container spacenet9:1.0 /bin/bash

Run Inference with Baseline

docker run -it --rm -v /test/data/:/test/data -v /path/to/this/repo:/path/to/this/repo --name inference spacenet9:1.0 python3 -m /path/to/this/repo/src/baseline/run.py --optical_image /test/data/optical_image.tif --sar_image /test/data/sar_image.tif --evaluation_tiepoints_filepath /test/data/tiepoints.csv --output_directory /test/data/my_output --log INFO

Workflow

1. Data Preparation

1. Reproject SAR and Optical image so that they are in the same coordinate reference system (see reproject_image() in alignment.py)
2. Align the SAR and Optical image (see alignment.py).
   • resample to lower resolution image
   • ensure same number of rows and columns
   • ensure same spatial origin
   • ensure same cell size

2. (Optional) Investigating baseline transformations on labeled tiepoints

Estimate a transformation using the given labeled tiepoints.

1. transform.py

3. Keypoint detection training

1. Generating training and validation tiles for training SAR and optical keypoint detection networks
   • Use generate_keypoint_patches.py and modify arguments.
   • Generate an opt_keypoint and sar_keypoint tiled dataset
2. Train SAR keypoint detection network
3. Train optical keypoint detection network

4. Running inference

1. run.py