RoboSat.pink

Semantic Segmentation ecosystem for GeoSpatial Imagery

This repository is a DataPink flavor of RoboSat, including our latests developments.

Spirit:

• Cutting edge Computer Vision research papers implementation
• Industrial code robustness
• Several tools, you can combine together (as Lego)
• Extensible, by design
• High performances
• Minimalism as a code aesthetic
• GeoSpatial standards compliancy
• OSM and MapBox ecosystems friendly
• PyTorch based

Aims:

• DataSet Quality Analysis
• Change Detection highlighter
• Features extraction and completion

Install:

1) Prerequisites:

• Python >= 3.6 and PyTorch >= 0.4 installed, with related Nvidia GPU drivers, CUDA and CUDNN libs.
• At least one GPU, with RAM GPU >= 6Go (default batch_size settings is targeted to 11Go).
• Libs with headers: libjpeg, libwebp, libbz2, zlib, libboost. And Qt dependancies: libsm and libxrender. On a recent Ubuntu-server, could be done with:

    apt-get install build-essential libboost-python-dev zlib1g-dev libbz2-dev libjpeg-turbo8-dev libwebp-dev libsm6 libxrender1

2) Python libs Install:

     python3 -m pip install -r requirements.txt

NOTA: if you want to significantly increase performances switch from Pillow to Pillow-simd.

3) Deploy:

• Move the rsp command to a bin directory covered by your PATH (or update your PATH)
• Move the robosat_pink dir to somewhere covered by your PYTHONPATH (or update your PYTHONPATH)

WorkFlows:

Related resources:

• RoboSat.pink tutorial: from OpenData to OpenDataSet
• RoboSat.pink documentation: Extensibility by Design
• Robosat slides @PyParis 2018
• MapBox RoboSat github directory
• Christoph Rieke's Awesome Satellite Imagery Datasets
• Mr Gloom's Awesome Semantic Segmentation

Bibliography:

• Optimizing IoU in Deep Neural Networks for Image Segmentation
• DeepRoadMapper: Extracting Road Topology from Aerial Images
• The Lovász-Softmax loss: A tractable surrogate for the optimization of the IoU measure in neural networks
• U-Net: Convolutional Networks for Biomedical Image Segmentation
• Deep Residual Learning for Image Recognition
• Angiodysplasia Detection and Localization Using Deep Convolutional Neural Networks
• TernausNetV2: Fully Convolutional Network for Instance Segmentation
• Joint Learning from Earth Observation and OpenStreetMap Data to Get Faster Better Semantic Maps
• In-Place Activated BatchNorm for Memory-Optimized Training of DNNs

Arch:

Contributions and Services:

• Pull Requests are welcome ! Feel free to send code... Don't hesitate either to initiate a prior discussion throught tickets on any implementation question.

• If you want to collaborate through code production and maintenance on a long term basis, please get in touch, co-edition with an ad hoc governance can be considered.

• If you want a new feature, but don't want to implement it, DataPink provide core-dev services.

• Expertise and training on RoboSat.pink are also provided by DataPink.

• And if you want to support the whole project, because it means for your own business, funding is also welcome.