Code for Change Event Dataset for Discovery from Spatio-temporal Remote Sensing Imagery

---

The code is divided 3 parts

1. Obtaining Sentinel-2 Data
2. Using Self-supervised Pipeline
3. Baselines

The code was tested on python 3.8. use requirements.txt. to install python.

pip install -r /path/to/requirements.txt.

---

Obtaining Sentinel-2 Data

Change the directory to data. To download the Raw input for CaiRoad use sentinel_cairo.py, for CalFire use sentinel_cali.py.

python3 sentinel_cairo.py

This code requires EarthEngine api, where an account needs to be created for authentication. Follow the from here for installation and authentication.

This also downloads the cloud masks for the dataset.

Note that the raw data obtained from this is already there on the project page. The directory is named fulldatain the downloaded CaiRoad.zip or CalFire.zip

Using Self-supervised Pipeline

Change the directory to pipeline. The pipeline itself has 3 parts:

Training and running pairwise change detection

To train our self-supervised change detection, use train_SSNet.py.

python3 train_SSNet.py -bf -fr -m train

The default batch size is set to 4. But on larger GPUs batch size can be increased. Instead of training, trained models can be downloaded from the project page.

For inferring use the same switches in the argument.

python3 train_SSNet.py -bf -fr -m inference

This will create a directory with binary change detection. The next step is grouping the changes.

Change grouping.

Use region_growing.py to group the changes.

python3 region_growing.py

This will create a directory with grouped change events. The changes are stored in two formats 1) a .npy file for storing the data and 2) png files fo visualization.

The change events are not yet in the standard format. Follow the next step for that.

Obtaining change events.

Use get_change_events.py to get individual change events from the change grouping.

python3 get_change_events.py

Alternatively, this data can also be downloaded from the project page.

The above mentioned pipeline can be used for any set of region to obtain summarizing change events from them.

---

Baselines

Change the directory to baselines. Use main.py to run different baselines. For example use the following for SimCLR:Change Events :

python3 main.py -dn CalFire -mt all -bs 256 -e 20

Change -dn to CaiRoad to run on CaiRoad dataset. Use arguments our of notime, nochange, fulldata, eurosat, imnet for -mt to run other baselines. The trained models are stored in a directory named models, that can used for inference. For inference use argument -m infer.

The reported results can be viewed using this.