32b1313d00f1a7000532fd191ad497aaac7bf8c4,imgaug/augmenters/arithmetic.py,Invert,_augment_images,#Invert#Any#Any#Any#Any#,1493

Before Change

                        image_c = image[..., c]
                        distance_from_min = np.abs(image_c - self.min_value) // d=abs(v-m)
                        image[..., c] = -distance_from_min + self.max_value // v"=M-d
                np.clip(image, 0, 255, out=image)
                result[i] = image.astype(np.uint8)
            else:
                p_sample = self.p.draw_sample(random_state=rs_image)
                assert 0 <= p_sample <= 1.0
                if p_sample > 0.5:
                    distance_from_min = np.abs(image - self.min_value) // d=abs(v-m)
                    image = -distance_from_min + self.max_value
                    np.clip(image, 0, 255, out=image)
                    result[i] = image.astype(np.uint8)
        return result

    def _augment_keypoints(self, keypoints_on_images, random_state, parents, hooks):
        return keypoints_on_images

After Change

        self.max_value = max_value

    def _augment_images(self, images, random_state, parents, hooks):
        input_dtypes = meta.copy_dtypes_for_restore(images)

        result = images
        nb_images = len(images)
        seeds = random_state.randint(0, 10**6, (nb_images,))
        for i in sm.xrange(nb_images):
            image = images[i].astype(np.int32)
            rs_image = ia.new_random_state(seeds[i])
            per_channel = self.per_channel.draw_sample(random_state=rs_image)
            if per_channel == 1:
                nb_channels = image.shape[2]
                p_samples = self.p.draw_samples((nb_channels,), random_state=rs_image)
                for c, p_sample in enumerate(p_samples):
                    assert 0 <= p_sample <= 1
                    if p_sample > 0.5:
                        image_c = image[..., c]
                        distance_from_min = np.abs(image_c - self.min_value) // d=abs(v-m)
                        image[..., c] = -distance_from_min + self.max_value // v"=M-d
                result[i] = image
            else:
                p_sample = self.p.draw_sample(random_state=rs_image)
                assert 0 <= p_sample <= 1.0
                if p_sample > 0.5:
                    distance_from_min = np.abs(image - self.min_value) // d=abs(v-m)
                    image = -distance_from_min + self.max_value
                    result[i] = image

        meta.clip_augmented_images_(result, self.min_value, self.max_value)
        meta.restore_augmented_images_dtypes_(result, input_dtypes)

        return result

    def _augment_keypoints(self, keypoints_on_images, random_state, parents, hooks):

Instances