Release 0.6.0

This release finally supports CUDA backends ✨
To use CUDA backends on multiple platforms please use the download from GoogleDrive:

• deepvision-cuda-0.6.0.zip [3.5GB]

Otherwise download the platform specific build from the attached downloads.

The following things are also included:

• Enabled support for Processing4 by removing AWT image to Mat conversion #7
• Reduced general file size by using javacpp platform builds
• Added support for arm and armhf platforms (raspberry pi)

Pre-Release 0.5.0

This release fixed a bug where this library is used together with the RealSense camera library for processing #5 .
It also adds some new networks but without examples yet:

• Midas Depth Estimation
• Open Face Vector Generator
• YOLOv4

Pre-Release 0.4.0

• This release updates the JavaCV backend to the SNAPSHOT of 1.5.3
• Adds a lot of new algorithms (TextBoxes, Tesseract, SuperResolution)

Pre-Release 0.3.4

• Fixes out-of-bounds bounding boxes for classifiers
• Added emotion detection through webcam example and facial landmark example
• Fixes YOLO Webcam example

Pre-Release 0.3.3

This release adds support for global and local installation as well as Gender and Age prediction algorithms.

Pre-Release 0.3.2

This release adds the support for RFB and slim 640 face detection networks, as well as a CNN based MNIST classifier.

Pre-Release 0.3.1

This release adds a new face detection algorithm which runs ultra fast on CPUs. Together with some adaptions for the single human pose detection, as well as some repository behaviours.

All the networks are now installed directly in sketch folder if DeepVision is initialised inside setup.

Pre-Release 0.2.0

This release adds support for inferencing a human pose estimation. At the moment the result is only used to detect one single human pose.

Pre-Release 0.1.0

Very early pre-release of the library which currently just supports inferencing YOLO algorithms.

Caution ⚡️: The dependencies are installed in your profile root directory. This will be changed in the next release.

Repository

Contains all the models and weights needed to run the algorithms. This data will be downloaded when setting up the network and resolving the dependencies.

• YOLO v3 Models and Weights by Joseph Redmon / pjreddie
• YOLO 9k Weights by Philippe Rémy / philipperemy
• YOLO v4 Models and Weights by AlexeyAB
• YOLO v5 Models by ultralytics
• Single Human Pose Estimation by Global Context for Convolutional Pose Machines
• Multi Human Pose Estimation by Daniil-Osokin/lightweight-human-pose-estimation
• Ultra-Light-Fast-Generic-Face-Detector-1MB by Linzaer
• MNIST by onnx/models
• FER+ Emotion Recognition by onnx/models
• SSR-Net for gender and age classification by shamangary
• Age and Gender Net by Gil Levi and Tal Hassner.
• FacemarkLBF Model by kurnianggoro/GSOC2017
• Handtrackjs Model by victordibia/handtracking
• COCO MobileNet by opencv
• Haar Cascade by OpenCV
• TextBoxes by MhLiao
• TextBoxes++ by MhLiao
• CRNN by bgshih
• Superresolution Models by Saafke
• YOLOv3 Hand Detector (Yolo & Tiny) trained by myself using CMU HandDB
• Cross Hands is trained by myself crossing 3 datasets (CMU, Egohands and my own)
• StyleTransfer models by eccv16
• YOLOv4 / YOLOv3 Tiny PRN / EfficientNetB0-Yolov3 / YOLOv3 OpenImage by AlexeyAB
• FaceEmeddingsExtraction by OpenFace
• Towards Fast, Accurate and Stable 3D Dense Face Alignment
• MiDas Net by Intel
• YOLO-Fastest dog-qiuqiu
• Colorful Image Colorization by Richard Zhang, Phillip Isola, Alexei A. Efros. In ECCV, 2016