LROC NAC MaskRCNN Prediction Pipeline

MaskRCNN prediction for LROC NAC images

Dependencies:

1. For running split_image_to_png.py, please install GDAL. (The code was run in GDAL version is 3.0.4)

   GDAL Python Installation Instructions

2. For running predict_lroc.py, please install NumPy, PyTorch, TorchVision and Pillow.

   PyTorch Installation Instructions

Models:

Download the model from Google Drive and save it in the models folder.

How to Run:

1. python3 split_image_to_png.py will split NAC_ROI_ALPHNSUSLOA_E129S3581_cropped.tif into 5 350x350 images and store them in predict_images directory.

2. python3 predict_lroc.py will predict on the images inside predict_images directory and output the corresponding PNG masks in predicted_masks folder.

Example output:

Convert the predictions to polygons:

gdal_polygonize.py predicted_masks/tile_350_0.png predicted_masks/tile_350_0.geojson -b 1 -f "GeoJSON" out DN

To create output prediction for input tif:

1. Copy all the *.xml files in predict_images to predicted_masks, and do a gdal_merge.

   gdal_merge.py -o NAC_ROI_ALPHNSUSLOA_E129S3581_predictions.tif predicted_masks/*.png