ofxOpenCvDnnSegmentation

Description

ofxOpenCvDnnSegmentation is an realtime segmentation addon for openframeworks, which uses opencv dnn modules.

OpenCV v.3.3.1 or upper includes some dnn modules in their own package. Therefore I designed ofxOpenCvDnnSegmentation with OpenCV v.3.3.1 or upper version. It works with ENET or FCN network.

Usage

See Examples for more details. You can find how to get each class detected segmentation with ofPolyline.

void setup()
{
    ofImage img;
    img.load(ofToDataPath("sample.jpg"));
    segmentation.setup(ofToDataPath("dnn/enet-model-best.net"),ofToDataPath("dnn/classlist.txt"));
    segmentation.update(img.getPixels());
}
void draw()
{
    ofBackground(0);
    segmentation.draw(0,0,ofGetWidth(), ofGetHeight());
}

Install

1. Clone ofxOpenCvDnnSegmentation to your of/addon directory

$ git clone https://github.com/TetsuakiBaba/ofxOpenCvDnnSegmentation.git

2. Download OpenCV.framework to ofxOpenCvDnnSegmentation/libs directory.

$ curl -O http://tetsuakibaba.jp/tmp/opencv2.framework.zip
$ unzip opencv2.framework.zip

Or you may build your own opencv.framework from opencv source. ( https://tetsuakibaba.jp/ws/doku.php?id=opencv_dnn:opencv2.framework )

Getting Started with Examples/single_image_or_video.

1. Download models

$ sh getModels.sh

2. Update single_image_or_video with the projectGenerator. Please be sure to include ofxOpenCvDnnSegmentation and ofxCv

3. Run

Compatibility

• only macOS ( tested on Mojave)
• of version: 0.10.1
• addon: ofxCv for OF master(0.10.0)

Sample Images

ENET results, Inference time: around 70[ms]

FCN Result. Inference time: around 1,600[ms]

Licence

• All materials except for the following videos and images:
  • MIT Lisence
• Videos, images
  • Video by Pixabay from Pexels,
  • Video by Gsm Tareq from Pexels

Author

TetsuakiBaba

Reference

1. https://docs.opencv.org/3.4/d4/d88/samples_2dnn_2segmentation_8cpp-example.html
2. ENET
3. FCN