Pipeline for training data creation, training and inference of a MaskRCNN to detect electyricxal substations in the Netherlands using NSO Superview - 0.5 m resolution

• conda: environment.yml

• Procedure torch, torchvision, detectron2

    pip3 install \
    torch==1.10.2 \
    torchvision==0.11.3 -extra-index-url https://download.pytorch.org/whl/cu113
    python -m pip install detectron2 -f \
    https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/index.html
  

• On cluster use:

    module load 2022
    module load CUDA/11.3.1
  

• Test environment by importing:

    import detectron2        
    import torch             
    import cv2 as cv         
    import numpy as np       
    from osgeo import gdal   
    from detectron2 import model_zoo
    from detectron2.engine import DefaultPredictor
    from detectron2.config import get_cfg
  

• Verify torch, torchvision, cuda compatibility by running:

    python -m detectron2.utils.collect_env
  
    See: https://stackoverflow.com/questions/70831932/cant-connect-to-gpu-when-building-pytorch-projects
    or
    Python -c "import uutils; uutils.torch_uu.gpu_test()
    see: https://stackoverflow.com/questions/66992585/how-does-one-use-pytorch-cuda-with-an-a100-gpu
  

• Aim at the preperation of the imagery and annotation for DL training

• download NSO rasters

• Extract and filter OSM data for electrical substations

• Create tiles of satelite imagery and annotation for DL model training

• Should be added to overide the gdal .ini file in conda env:

    osmconf.ini
  

• Also in subs_detection/scripts/extract_osm_sub.py:

    gdal.SetConfigOption("OSM_CONFIG_FILE", os.path.join('..',"osmconf.ini"))"