How to Create A Component

It is recommended to create a component using templates instead of starting from scratch. Go to browser and click on Create new component, select python Crop Image template. This will generate an example component for cropping an image. It is recommended to modify it rather than create everything from scratch.

It is possible to use PPL Tool for initializing and creating more complicated components.

Steps

Component Structure

This is should be your folder structure after you create component with template

Structure for Dot Product template

.
├── component.yml
├── .pipecomponent
├── src
│   ├── requirements.txt
│   └── main.py
└── tests
    ├── test_0
    │   ├── input_0.txt
    │   ├── input_1.txt
    │   └── output_0.txt
    └── test_1
        ├── input_0.txt
        ├── input_1.txt
        └── output_0.txt

Structure for Crop Image template

.
├── component.yml
├── .pipecomponent
├── src
   ├── requirements.txt
   └── main.py

Configuration File

name: "Crop Image Python"
language: py
platform: linux
build_system: 1.1.3
worker:
  input_type: "Image"
  output_type: "Image"
	config_schema:
  	top_left:
  	type: (UInt64, UInt64)
  	bottom_right:
  	type: (UInt64, UInt64)

Crop Image Source

from pipelogic.worker import run, config
import pipelogic
import numpy as np


def crop_image(Image):
    raw_data = np.reshape(np.array(Image.data, dtype='uint8'), (Image.height, Image.width, 3))
    x1 = config.top_left[0]
    y1 = config.top_left[1]
    x2 = config.bottom_right[0]
    y2 = config.bottom_right[1]

    image_cropped = raw_data[x1:x2, y1:y2]

    return { "height": image_cropped.shape[0],
             "width": image_cropped.shape[1],
             "data": image_cropped.flatten().tobytes(),
             "format": Image.format}

run(crop_image)

Rewrite the component

Convert Image to Gray will process images and will output images. Therefore, the input and output types stay same. We need to remove config_schema section since it does not require any parameter. We also nned to change the name of the component.

Configuration

name: "Convert Image to Gray"
language: py
platform: linux
build_system: 1.1.3
tags: ["one-to-one", "image-manipulation"]
worker:
  input_type: "Image"
	output_type: "Image"

Dependencies

Inside requirements.txt, first library provided pipelogic api and the other ones are needed for image manipulation.

pipelogic
numpy
opencv-python-headless # headless version is for servers

Component function

Add libraries to source code and change the source code by replacing crop_image with convert_gray.

from pipelogic.worker import run, config
import pipelogic
import numpy as np
import cppipe
import cv2

Image_GRAY = cppipe.create_named('Image.GRAY', cppipe.create_tuple([]))

def convert_gray(Image):
    image = np.reshape(np.array(Image.data, dtype='uint8'), (Image.height, Image.width, 3))
    gray_image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)

    return {
        "height": Image.height,
        "width": Image.width,
        "data": gray_image.flatten().tobytes(),
        "format": pipelogic.types.Named(pipelogic.types.Tuple(()), Image_GRAY)
    }

run(convert_gray)

Extend Color Conversion

Add additional format definitions and a function to make component work for every channel format.

from pipelogic.worker import run, config
import pipelogic
import numpy as np
import cppipe
import cv2

Image_BGR = cppipe.create_named('Image.BGR', cppipe.create_tuple([]))
Image_RGB = cppipe.create_named('Image.RGB', cppipe.create_tuple([]))
Image_GRAY = cppipe.create_named('Image.GRAY', cppipe.create_tuple([]))
Image_RGBA = cppipe.create_named('Image.RGBA', cppipe.create_tuple([]))
Image_BGRA = cppipe.create_named('Image.BGRA', cppipe.create_tuple([]))

# converts all possible image formats into gray image
def gray_conversion(Image):
    image_data = np.array(Image.data, dtype='uint8')
    if Image.format == Image_BGR:
    	 image = np.reshape(image_data, (Image.height, Image.width, 3))
         return cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
    elif Image.format == Image_RGB:
    	 image = np.reshape(image_data, (Image.height, Image.width, 3))
         return cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
    elif Image.format == Image_RGBA:
    	 image = np.reshape(image_data, (Image.height, Image.width, 4))
         return cv2.cvtColor(image, cv2.COLOR_RGBA2GRAY)
    elif Image.format == Image_BGRA:
    	 image = np.reshape(image_data, (Image.height, Image.width, 4))
         return cv2.cvtColor(image, cv2.COLOR_BGRA2GRAY)
    else:
    	 image = np.reshape(image_data, (Image.height, Image.width, 1))
         return image	   

def convert_gray(Image):
    gray_image = gray_conversion(Image)

    return {
       "height": Image.height,
         "width": Image.width,
         "data": gray_image.flatten().tobytes(),
         "format": pipelogic.types.Named(pipelogic.types.Tuple(()), Image_GRAY)
    }

run(convert_gray)