PictureFlow

PictureFlow is an image processing library that aims to simplify the way we think about modifying images. It depends on OpenCV for most image manipulation tasks, altough we have plans to support more backends in the future. We currently only support Python 3.6+, but efforts to make the project compatible with 2.7+ and < 3.6 are very welcome!

Installation

Via pip

$ pip install pictureflow

Manually

$ git clone git@github.com:mentum/pictureflow.git
$ cd pictureflow
$ python setup.py install

Example

from pictureflow.output import DiskOutput
from pictureflow.transform import ColorMask, Convert, Rotate, Scale

import cv2
import pictureflow as pf

# Define some placeholders that will form the root of our transformation graph
imgs = pf.Placeholder(out_type=pf.Image)
angle = pf.Placeholder(out_type=int)

# This branch will reduce the image to 20% of its original size, convert it to RGB & write it to the 'output/' directory
branch_a = Scale(imgs, scale_factor=pf.Constant(0.2))
branch_a = Convert(branch_a, src=pf.Constant('bgr'), dest=pf.Constant('rgb'))
branch_a = DiskOutput(branch_a, base_path=pf.Constant('output/'))

# This branch will rotate the image by variable angles (fed with the session context),
# will filter-out the red color, and will write it to the 'other/' directory.
branch_b = Rotate(imgs, rot_angle=angle)
branch_b = ColorMask(branch_b, color_mask=pf.Constant([None, None, 0]))
branch_b = DiskOutput(branch_b, base_path=pf.Constant('other/'))

# Load images
image_1 = pf.Image('image_1', 'jpg', cv2.imread('pic.jpg'))
image_2 = pf.Image('image_2', 'jpg', cv2.imread('pic2.jpg'))

# Declare a new session. (A session is basically a runtime context for a
# previously-defined graph)
with pf.Session({imgs: [image_1, image_2], angle: [90, 180]}) as sess:
    output = sess.run_sync(branch_a)

    for item in sess.run(branch_b):
        print(item.id)

Requirements

• Python 3.6