Visualizer

visualize_array(array, out_path=None)

Simple wrapper function to visualize a NumPy matrix as image via PIL.

Info

NumPy array shape has to be (x, y, channel) like this: (224, 224, 3)

Parameters:

Name	Type	Description	Default
array	numpy.ndarray	NumPy matrix containing an image.	required
out_path	str	Path in which image is stored (else live output).	None

Source code in aucmedi/utils/visualizer.py

def visualize_array(array, out_path=None):
    """ Simple wrapper function to visualize a NumPy matrix as image via PIL.

    ???+ info
        NumPy array shape has to be (x, y, channel) like this: (224, 224, 3)

    Args:
        array (numpy.ndarray):          NumPy matrix containing an image.
        out_path (str):                 Path in which image is stored (else live output).
    """
    # Ensure integer intensity values
    array = np.uint8(array)
    # Remove channel axis if grayscale
    if array.shape[-1] == 1 : array = np.reshape(array, array.shape[:-1])
    # Convert array to PIL image
    image = Image.fromarray(array)
    # Visualize or store image
    if out_path is None : image.show()
    else : image.save(out_path)

visualize_heatmap(image, heatmap, out_path=None, alpha=0.4)

Simple wrapper function to visualize a heatmap encoded as NumPy matrix with a [0-1] range as image via matplotlib and PILLOW.

Reference - Implementation

Author: François Chollet
Date: April 26, 2020
https://keras.io/examples/vision/grad_cam/

Parameters:

Name	Type	Description	Default
image	numpy.ndarray	NumPy matrix containing an image.	required
heatmap	numpy.ndarray	NumPy matrix containing a XAI heatmap.	required
out_path	str	Path in which image is stored (else live output).	None
alpha	float	Transparency value for heatmap overlap on image (range: [0-1]).	0.4

Source code in aucmedi/utils/visualizer.py

def visualize_heatmap(image, heatmap, out_path=None, alpha=0.4):
    """ Simple wrapper function to visualize a heatmap encoded as NumPy matrix with a
        [0-1] range as image via matplotlib and PILLOW.

    ??? abstract "Reference - Implementation"
        Author: François Chollet <br>
        Date: April 26, 2020 <br>
        https://keras.io/examples/vision/grad_cam/ <br>

    Args:
        image (numpy.ndarray):          NumPy matrix containing an image.
        heatmap (numpy.ndarray):        NumPy matrix containing a XAI heatmap.
        out_path (str):                 Path in which image is stored (else live output).
        alpha (float):                  Transparency value for heatmap overlap on image (range: [0-1]).
    """
    # If image is grayscale, convert to RGB
    if image.shape[-1] == 1 : image = np.concatenate((image,)*3, axis=-1)
    # Rescale heatmap to grayscale range
    heatmap = np.uint8(heatmap * 255)
    # Use jet colormap to colorize heatmap
    jet = cm.get_cmap("jet")
    # Use RGB values of the colormap
    jet_colors = jet(np.arange(256))[:,:3]
    jet_heatmap = jet_colors[heatmap] * 255
    # Superimpose the heatmap on original image
    si_img = jet_heatmap * alpha + (1-alpha) * image
    # Convert array to PIL image
    si_img = si_img.astype(np.uint8)
    pil_img = Image.fromarray(si_img)
    # Visualize or store image
    if out_path is None : pil_img.show()
    else : pil_img.save(out_path)