/detectron2-spacenet

A repository that hosts installation scripts for an accompanying Medium article. These scripts will download the images and prepare them for the Detectron2 algorithm.

Primary LanguagePython

Start here.

  1. Clone this repository. 
    # HTTPS
git clone https://github.com/rl02898/detectron2-spacenet.git
# SSH
git clone git@github.com:rl02898/detectron2-spacenet.git
  2. Run the following command. This command will download the data, put it into directories, rename it, convert the TIFF images to PNGs, and create the JSON files for Detectron2. This will take an hour or two. You may see warnings, they are expected. 
    ./install.sh
    with a file structure like: 
    Spacenet/
├─ AOI_1_Rio_train/
...
...
...
├─ AOI_5_Khartoum_Test_public/
├─ train/
│  ├─ images.png
│  ├─ via_region_data.json
├─ val/
│  ├─ images.png
│  ├─ via_region_data.json
├─ test/
│  ├─ images.png
  3. Follow the steps in the Medium article on how to properly use the inference functions in demo.py:
    https://russland.medium.com/using-detectron2-for-instance-segmentation-on-the-spacenet-dataset-94338f739cd0