Port Computer Vision Model to Android

- Freelancer must have demonstrable previous experience in porting computer vision or machine learning-based algorithms in python to java or C++ - Ideally freelancer will have ported pytorch code TensorFlow / TFLite - Freelancer must deliver by Thursday 30th July. -- Port our Social Distancing algorithm (written in Python/Pytorch) to Java or C++ to make compatible with AndroidOS so it can process videos (with similar speed and accuracy to its current state) on-device (an edge-enabled security camera). It will then be uploaded as an app on the Safety & Security Things GmbH marketplace. Access to our GitHub repo with our social distancing code (written in python/pytorch) will be provided Security & Safety Things (S&ST) provide the necessary documentation to support the porting: [login to view URL] (login info provided) Some of the key points can be summarised below: OS: Android Open Source Project (AOSP). It includes S&ST APIs and OS components for managing video pipelines, web server and UI, user db, message broker etc. SUPPORTED PROGRAMMING LANGUAGES: - Java/Kotlin - C/C++ SUPPORTED INFERENCE FRAMEWORKS: - TensorFlow Lite - Qualcomm Snapdragon Neural Processing Engine (SNPE) In order to run on the app existing models must be converted - from TensorFlow model to TensorFlow Lite format using TensorFlow tools (e.g. tflite_convert) - from Caffe2, TensorFlow and ONNX to Qualcomm Deep Learning Container (DLC) using the Qualcomm SNPE SDK *Although the provided documentation doesn’t mention pytorch it is possible to convert a pytorch model into a format that is either supported (pytorch -> tensorflow lite) or can be ported (pytorch -> ONNX) ([login to view URL])