• http://xviewdataset.org
• https://github.com/DIUx-xView
• https://challenge.xviewdataset.org/baseline
• https://challenge.xviewdataset.org/tutorial
• https://github.com/ultralytics/xview-docker
• https://github.com/PlatformStories/train-cnn-classifier
• https://medium.com/picterra/the-xview-dataset-and-baseline-results-5ab4a1d0f47f

• https://github.com/databricks/spark-deep-learning
• https://github.com/yahoo/TensorFlowOnSpark
• https://towardsdatascience.com/deep-learning-with-apache-spark-part-2-2a2938a36d35
• https://github.com/cloudmesh/example-project-nist-fingerprint-matching
• https://medium.com/linagora-engineering/making-image-classification-simple-with-spark-deep-learning-f654a8b876b8
• oshinko
  • https://github.com/radanalyticsio
  • https://www.youtube.com/watch?v=IrHTeco4r_Q

• red blood cell detection
  • https://gist.github.com/saudet/fb8a4d9544dc3c411b302ccd6bbf87e7
  • https://github.com/cosmicad/dataset
• SIMRDWN
  • https://github.com/CosmiQ/simrdwn
  • https://medium.com/the-downlinq/simrdwn-adapting-multiple-object-detection-frameworks-for-satellite-imagery-applications-991dbf3d022b
• YOLT
  • https://github.com/CosmiQ/yolt
  • https://medium.com/the-downlinq/you-only-look-twice-multi-scale-object-detection-in-satellite-imagery-with-convolutional-neural-38dad1cf7571
  • https://medium.com/the-downlinq/yolt-arxiv-paper-and-code-release-8b30d40d095b
• http://fxapps.blogspot.com/2018/01/detecting-objects-using-javafx-and-deep.html
• https://github.com/ultralytics
• https://playground.tensorflow.org
• https://blog.skymind.ai/building-a-production-grade-object-detection-system-with-skil-and-yolo/
• https://skymind.ai/wiki/convolutional-network

• https://github.com/DIUx-xView/baseline/releases
• https://www.nist.gov/itl/iad/image-group/resources/biometric-special-databases-and-software
• https://github.com/cosmicad/dataset
• http://www.cvlibs.net/datasets/kitti/
• https://geodacenter.github.io/data-and-lab/

see labels.txt

• install nvidia-docker
  • https://github.com/NVIDIA/nvidia-docker/wiki/Installation-(version-2.0)#prerequisites
• install cuda
  • https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html
  • https://developer.nvidia.com/cuda-downloads
• register for nvidia registry
  • https://ngc.nvidia.com/registry
• run containers with nvidia-docker
  • nvidia-docker run --rm nvidia/cuda nvidia-smi
• debug issues with
  • nvidia-container-cli -k -d /dev/tty info

The json is easier to work with when partitioned

jq '[.features[] | select(.properties.image_id == "100.tif")]' xView_train.geojson

for f in *.tif; do echo $f; jq --arg f "$f" '[.features[] | select(.properties.image_id == $f)]' ../xView_train.geojson > $f.geojson; done

• https://developer.nvidia.com/digits
• https://devblogs.nvidia.com/deep-learning-object-detection-digits/

Have to adjust the boto library configuration to use the non-subdomain s3 urls

Mount this as /etc/boto.cfg

[s3]
calling_format=boto.s3.connection.OrdinaryCallingFormat

• NVIDIA/DIGITS#992
• https://github.com/NVIDIA/DIGITS/blob/master/docs/ImageFolderFormat.md
  • PNG, JPEG, BMP and PPM