ImageComposition

Image Composition Algorithm to make to make train data when there is not enough train images.

This Algorithm is basicly for YOLACT model (Real-Time Instance Segmentation Model)

YOCLACT

If you don't have enough train data to train, you might use this algorithm to generate new train images using existing train images.

You need Json File which can be generated from 'Labelme'

Labelme

  json = {}
  json['version'] = "4.1.1"
  json['flags'] = {}
  json['shapes'] = []
  json['imagePath'] = fname
  json['imageData'] = imagedata.imagedata(path).decode('utf-8')
  json['imageHeight'] = height
  json['imageWidth'] = width
  json['lineColor'] = [0,255,0,128]
  json['fillColor'] = [255,0,0,128]

Environment

---

• Python == 3.6x
• Set up the environment using one of the following methods:

conda install pillow
conda install opencv

RGBA2RGB

---

Sometimes if you want to use png file, you may need to change RGBA to RGB Especially when you making train set for YOLACT.

• You can use

RGBA2RGB.py