/DNNAssist

Simple Qt/Qml application for training an validating various DNN frameworks models against media

Primary LanguageMakefileMIT LicenseMIT

DNNAssist!

DNNAssist is a simple Qt/Qml application for testing various DNN (deep neural networks) frameworks.
a

IDEA

Load trained model and easy manipulating visual media and parameters of detection while WSIWYG (what you see is what you get) and compare them between various networks.

Choose media

Switch between local media:

  • Image file
  • Movie file
  • Device camera.

Note: In fact of a unresolved yet bug with media plugin on linux is not posible to switch from camera to RTSP or youtube and switch camera again. It wil be resolved further. At windows works perfectlly.

and network based ones:

  • link to image
  • RTSP stream, i.e. you own IP camera, phone with streaming application, public IP cameras

Note: there can be additional lag in transmission so live processing can be not reachable.

  • youtube clip

Note: given youtube link is send to external server to resolve exact clip location. Due limitation of API not all clips can be processed. If pasted une is not working, use different one.

Loading trained models

a

You can your custom trained models or use pretrained one. Currentlly supported one are:

  • caffe
  • tensorflow
  • yolo

SUPPORTED FRAMEWORKS & MODELS

a Currentlly dnn processing is done only by OpenCv. In futher released additional framework will be added (i.e. Darknet, OpenVino, Torch and more). Currentlly supported dnn frameworks models are:

  • Tensorflow, YOLO - supported any custom or pretrained object recognition models also with masks (i.e. MaskRCNN).
  • Caffe - supported any custom or pretrained object recognition models. Masks are not currentlly impelemnted.

Other frameworks and models will be handled in future releases

HOW TO BUILD

Build OpenCV for desired platform

  • Linux - depends on you hardware build OpenCV with or without GPU support i.e. using this link.
  • Windows - to build OpenCv with GPU support use MSCV for both, OpenCv and DNNAssist. OpenCv do not support MinGW on windows.

Build or install OpenSSL

  • Windows - https://doc.qt.io/qt-5/windows-requirements.html
  • Linux - sudo apt-get install libssl-dev

Open issues: