Feature Request for GammaAdjustment Augmentation

[sparkfish]

Feature Request for GammaAdjustment Augmentation

[shaheryar1]

Got it. I guess this issue can occur in any of the augmentation type mismatch. There should be a away to restrict this

[kwcckw]

gemma

Sorry but do you mean gamma? Gamma correction with Power Law transformation should be useful in adjusting the image intensity and hence giving a better visualization.

[shaheryar1]

Yes Gamma. Sorry for typos. I will work on my next PR to add this feature in repo

[jboarman]

I went ahead and merged this in since it looks pretty straightforward. I'll hold off on publishing to PyPI until someone else has had a chance to test it out.

[kwcckw]

I tested it and looks fine, but one thing to take note is this augmentation is only working at post phase, and we will getting error if we put it in paper or ink phase, so it is better to add some notes and code to avoid the error if user put it at ink or paper phase. For example, adding evaluation of post data:

   class GammaAugmentation(Augmentation):
      def __init__(self, range=(0.5, 1.5), probability=0.5):
        super().__init__(probability=probability)
        self.range = range
    
      def __repr__(self):
        return f"GammaAugmentation(range={self.range}, probability={self.probability})"
    
      def __call__(self, data,force=False):
        if data['post']:
            img = data['post'][-1].result
            img = img.astype(np.uint8)
            value = random.uniform(self.range[0], self.range[1])
            invGamma = 1.0 / value
            table = np.array([((i / 255.0) ** invGamma) * 255
                              for i in np.arange(0, 256)]).astype("uint8")
            print(table.shape,img.shape)
            frame= cv2.LUT(img, table)
            data['post'].append(AugmentationResult(self, frame))

Example with gamma of 0.5 amd 2: