edafa

Test Time Augmentation (TTA) wrapper for computer vision tasks: segmentation,classification, super-resolution, ... etc.


Keywords
augmentation classification segmentation super-resolution pansharpening keras tensorflow pytorch, augmentation, classification, computer-vision, keras, pytorch, segmentation, super-resolution, tensorflow, wrapper
License
MIT
Install
pip install edafa==0.1.4

Documentation

Edafa

GitHub contributions welcome HitCount
Edafa is a simple wrapper that implements Test Time Augmentations (TTA) on images for computer vision problems like: segmentation, classification, super-resolution, Pansharpening, etc. TTAs guarantees better results in most of the tasks.

Test Time Augmentation (TTA)

Applying different transformations to test images and then average for more robust results.

pipeline

Installation

pip install edafa

Getting started

The easiest way to get up and running is to follow example notebooks for segmentation and classification showing TTA effect on performance.

How to use Edafa

The whole process can be done in 4 steps:

  1. Import Predictor class based on your task category: Segmentation (SegPredictor) or Classification (ClassPredictor)
from edafa import SegPredictor
  1. Inherit Predictor class and implement the main function
    • predict_patches(self,patches) : where your model takes image patches (numpy.ndarray) and return prediction (numpy.ndarray)
class myPredictor(SegPredictor):
    def __init__(self,model,*args,**kwargs):
        super().__init__(*args,**kwargs)
        self.model = model

    def predict_patches(self,patches):
        return self.model.predict(patches)
  1. Create an instance of you class
p = myPredictor(model,patch_size,model_output_channels,conf_file_path)
  1. Call predict_images() to run the prediction process
p.predict_images(images,overlap=0)

Configuration file

Configuration file is a json file containing two pieces of information

  1. Augmentations to apply (augs). Supported augmentations:
    • NO : No augmentation
    • ROT90 : Rotate 90 degrees
    • ROT180 : Rotate 180 degrees
    • ROT270 : Rotate 270 degrees
    • FLIP_UD : Flip upside-down
    • FLIP_LR : Flip left-right
    • BRIGHT : Change image brightness randomly
    • CONTRAST : Change image contrast randomly
    • GAUSSIAN : Add random gaussian noise
    • GAMMA : Perform gamma correction with random gamma
  2. Combination of the results (mean). Supported mean types:
    • ARITH : Arithmetic mean
    • GEO : Geometric mean
  3. Number of bits image (default is 8-bits) (bits).

Example of a conf file

{
"augs":["NO",
"FLIP_UD",
"FLIP_LR"],
"mean":"ARITH",
"bits":8
}

You can either pass file path or the actual json text to conf parameter.

Contribution

All contributions are welcomed. Please make sure that all tests passed before pull request. To run tests

nosetests