albumentations-demo / README.md
IliaLarchenko's picture
Merged professional mode to master
29137d1
|
raw
history blame
2.9 kB

albumentations-demo

This service is created to demonstrate abilities of the Albumentations - a library for efficient image augmentations. Link to my article about augmentations selection and why this service can be useful

Easy start

You can play with this service right now https://albumentations-demo.herokuapp.com/ (It is deployed on free service with very limited computing power and can be quite unstable)

If you would like to run it locally follow the installation instruction.

Installation

git clone https://github.com/IliaLarchenko/albumentations-demo
cd albumentations-demo
pip install -r requirements.txt
streamlit run src/app.py

If you want to work with you own images just replace the last line with:

streamlit run src/app.py -- --image_folder PATH_TO_YOUR_IMAGE_FOLDER

If your images have some unusual proportions you can use image_width parameter to set the width in pixels of the original image to show. The width of the transformed image and heights of both images will be computed automatically. Default value of width is 400.

streamlit run src/app.py -- --image_width INT_VALUE_OF_WIDTH

In your terminal you will see the link to the running local service similar to :

  You can now view your Streamlit app in your browser.

  Network URL: http://YOUR_LOCAL_IP:8501
  External URL: http://YOUR_GLOBAL_IP:8501

Just follow the local link to use the service.

How to use

The interface is very simple and intuitive:

  1. On the left you have a control sidebar. Select the "Simple" mode. You can choose the image and the transformation.
  2. After that you will see the control elements for the every parameter this transformation has.
  3. Every time you change any parameter you will see the augmented version of the image on the right side of your screen.
  4. Below the images you can find a code for calling of the augmentation with selected parameters.
  5. You can also find there the original docstring for this transformation. screenshot

Professional mode

In the professional mode you can:

  1. Upload your own image
  2. Combine multiple transformations
  3. See the random parameters used to get the result

Be aware that in Professional mode some combination of parameters of different transformations can be invalid. You should control it.

Links