experiments/overlap/augmentations/pil.py [480:519]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    def sample_parameters(self):
        offset_x = float_parameter(sample_level(self.severity,self.max_intensity), 0.1)
        if np.random.uniform() > 0.5:
            offset_x = -offset_x
        offset_y = float_parameter(sample_level(self.severity,self.max_intensity), 0.1)
        if np.random.uniform() > 0.5:
            offset_y = -offset_y
        shift_x = float_parameter(sample_level(self.severity,self.max_intensity), self.im_size / 10)
        #shift_x = 0.0
        if np.random.uniform() > 0.5:
            shift_x = -shift_x
        shift_y = float_parameter(sample_level(self.severity,self.max_intensity), self.im_size / 10)
        #shift_y = 0.0
        if np.random.uniform() > 0.5:
            shift_y = -shift_y
        factor_x = float_parameter(sample_level(self.severity,self.max_intensity), 0.15)
        if np.random.uniform() > 0.5:
            factor_x = -factor_x
        factor_x = 2 ** factor_x
        factor_y = float_parameter(sample_level(self.severity,self.max_intensity), 0.15)
        if np.random.uniform() > 0.5:
            factor_y = -factor_y
        factor_y = 2 ** factor_y
        denom_x = float_parameter(sample_level(self.severity,self.max_intensity), 0.2 / self.im_size)
        if np.random.uniform() > 0.5:
            denom_x = denom_x
        denom_y = float_parameter(sample_level(self.severity,self.max_intensity), 0.2 / self.im_size)
        if np.random.uniform() > 0.5:
            denom_y = denom_y
        perspective_params = np.array([factor_x, offset_x, shift_x,offset_y, factor_y, shift_y, denom_x, denom_y])
        return {'perspective_params' : perspective_params}

    def transform(self, image, perspective_params):
        im = Image.fromarray(image)
        im = im.transform(
                (self.im_size, self.im_size), 
                Image.PERSPECTIVE,
                perspective_params, 
                resample=Image.BILINEAR
                )
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



experiments/overlap/augmentations/pil.py [535:574]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
    def sample_parameters(self):
        offset_x = float_parameter(sample_level(self.severity,self.max_intensity), 0.1)
        if np.random.uniform() > 0.5:
            offset_x = -offset_x
        offset_y = float_parameter(sample_level(self.severity,self.max_intensity), 0.1)
        if np.random.uniform() > 0.5:
            offset_y = -offset_y
        shift_x = float_parameter(sample_level(self.severity,self.max_intensity), self.im_size / 10)
        #shift_x = 0.0
        if np.random.uniform() > 0.5:
            shift_x = -shift_x
        shift_y = float_parameter(sample_level(self.severity,self.max_intensity), self.im_size / 10)
        #shift_y = 0.0
        if np.random.uniform() > 0.5:
            shift_y = -shift_y
        factor_x = float_parameter(sample_level(self.severity,self.max_intensity), 0.15)
        if np.random.uniform() > 0.5:
            factor_x = -factor_x
        factor_x = 2 ** factor_x
        factor_y = float_parameter(sample_level(self.severity,self.max_intensity), 0.15)
        if np.random.uniform() > 0.5:
            factor_y = -factor_y
        factor_y = 2 ** factor_y
        denom_x = float_parameter(sample_level(self.severity,self.max_intensity), 0.2 / self.im_size)
        if np.random.uniform() > 0.5:
            denom_x = denom_x
        denom_y = float_parameter(sample_level(self.severity,self.max_intensity), 0.2 / self.im_size)
        if np.random.uniform() > 0.5:
            denom_y = denom_y
        perspective_params = np.array([factor_x, offset_x, shift_x,offset_y, factor_y, shift_y, denom_x, denom_y])
        return {'perspective_params' : perspective_params}

    def transform(self, image, perspective_params):
        im = Image.fromarray(image)
        im = im.transform(
                (self.im_size, self.im_size), 
                Image.PERSPECTIVE,
                perspective_params, 
                resample=Image.BILINEAR
                )
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -