Bases: XAImethod_Base
XAI Method for Saliency Map (also called Backpropagation).
Normally, this class is used internally in the aucmedi.xai.decoder.xai_decoder in the AUCMEDI XAI module.
Reference - Implementation #1
Author: Yasuhiro Kubota
GitHub Profile: https://github.com/keisen
Date: Aug 11, 2020
https://github.com/keisen/tf-keras-vis/
Reference - Implementation #2
Author: Huynh Ngoc Anh
GitHub Profile: https://github.com/experiencor
Date: Jun 23, 2017
https://github.com/experiencor/deep-viz-keras/
Reference - Publication
Karen Simonyan, Andrea Vedaldi, Andrew Zisserman. 20 Dec 2013.
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps.
https://arxiv.org/abs/1312.6034
This class provides functionality for running the compute_heatmap function,
which computes a Saliency Map for an image with a model.
Source code in aucmedi/xai/methods/saliency.py
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107 | class SaliencyMap(XAImethod_Base):
""" XAI Method for Saliency Map (also called Backpropagation).
Normally, this class is used internally in the [aucmedi.xai.decoder.xai_decoder][] in the AUCMEDI XAI module.
??? abstract "Reference - Implementation #1"
Author: Yasuhiro Kubota <br>
GitHub Profile: [https://github.com/keisen](https://github.com/keisen) <br>
Date: Aug 11, 2020 <br>
[https://github.com/keisen/tf-keras-vis/](https://github.com/keisen/tf-keras-vis/) <br>
??? abstract "Reference - Implementation #2"
Author: Huynh Ngoc Anh <br>
GitHub Profile: [https://github.com/experiencor](https://github.com/experiencor) <br>
Date: Jun 23, 2017 <br>
[https://github.com/experiencor/deep-viz-keras/](https://github.com/experiencor/deep-viz-keras/) <br>
??? abstract "Reference - Publication"
Karen Simonyan, Andrea Vedaldi, Andrew Zisserman. 20 Dec 2013.
Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps.
<br>
[https://arxiv.org/abs/1312.6034](https://arxiv.org/abs/1312.6034)
This class provides functionality for running the compute_heatmap function,
which computes a Saliency Map for an image with a model.
"""
def __init__(self, model, layerName=None):
""" Initialization function for creating a Saliency Map as XAI Method object.
Args:
model (keras.model): Keras model object.
layerName (str): Not required in Saliency Maps, but defined by Abstract Base Class.
"""
# Cache class parameters
self.model = model
#---------------------------------------------#
# Heatmap Computation #
#---------------------------------------------#
def compute_heatmap(self, image, class_index, eps=1e-8):
""" Core function for computing the Saliency Map for a provided image and for specific classification outcome.
???+ attention
Be aware that the image has to be provided in batch format.
Args:
image (numpy.ndarray): Image matrix encoded as NumPy Array (provided as one-element batch).
class_index (int): Classification index for which the heatmap should be computed.
eps (float): Epsilon for rounding.
The returned heatmap is encoded within a range of [0,1]
???+ attention
The shape of the returned heatmap is 2D -> batch and channel axis will be removed.
Returns:
heatmap (numpy.ndarray): Computed Saliency Map for provided image.
"""
# Compute gradient for desierd class index
with tf.GradientTape() as tape:
inputs = tf.cast(image, tf.float32)
tape.watch(inputs)
preds = self.model(inputs)
loss = preds[:, class_index]
gradient = tape.gradient(loss, inputs)
# Obtain maximum gradient based on feature map of last conv layer
gradient = tf.reduce_max(gradient, axis=-1)
# Convert to NumPy & Remove batch axis
heatmap = gradient.numpy()[0]
# Intensity normalization to [0,1]
numer = heatmap - np.min(heatmap)
denom = (heatmap.max() - heatmap.min()) + eps
heatmap = numer / denom
# Return the resulting heatmap
return heatmap
|
__init__(model, layerName=None)
Initialization function for creating a Saliency Map as XAI Method object.
Parameters:
| Name |
Type |
Description |
Default |
model |
keras.model
|
Keras model object. |
required
|
layerName |
str
|
Not required in Saliency Maps, but defined by Abstract Base Class. |
None
|
Source code in aucmedi/xai/methods/saliency.py
57
58
59
60
61
62
63
64
65 | def __init__(self, model, layerName=None):
""" Initialization function for creating a Saliency Map as XAI Method object.
Args:
model (keras.model): Keras model object.
layerName (str): Not required in Saliency Maps, but defined by Abstract Base Class.
"""
# Cache class parameters
self.model = model
|
compute_heatmap(image, class_index, eps=1e-08)
Core function for computing the Saliency Map for a provided image and for specific classification outcome.
Attention
Be aware that the image has to be provided in batch format.
Parameters:
| Name |
Type |
Description |
Default |
image |
numpy.ndarray
|
Image matrix encoded as NumPy Array (provided as one-element batch). |
required
|
class_index |
int
|
Classification index for which the heatmap should be computed. |
required
|
eps |
float
|
Epsilon for rounding. |
1e-08
|
The returned heatmap is encoded within a range of [0,1]
Attention
The shape of the returned heatmap is 2D -> batch and channel axis will be removed.
Returns:
| Name | Type |
Description |
heatmap |
numpy.ndarray
|
Computed Saliency Map for provided image. |
Source code in aucmedi/xai/methods/saliency.py
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107 | def compute_heatmap(self, image, class_index, eps=1e-8):
""" Core function for computing the Saliency Map for a provided image and for specific classification outcome.
???+ attention
Be aware that the image has to be provided in batch format.
Args:
image (numpy.ndarray): Image matrix encoded as NumPy Array (provided as one-element batch).
class_index (int): Classification index for which the heatmap should be computed.
eps (float): Epsilon for rounding.
The returned heatmap is encoded within a range of [0,1]
???+ attention
The shape of the returned heatmap is 2D -> batch and channel axis will be removed.
Returns:
heatmap (numpy.ndarray): Computed Saliency Map for provided image.
"""
# Compute gradient for desierd class index
with tf.GradientTape() as tape:
inputs = tf.cast(image, tf.float32)
tape.watch(inputs)
preds = self.model(inputs)
loss = preds[:, class_index]
gradient = tape.gradient(loss, inputs)
# Obtain maximum gradient based on feature map of last conv layer
gradient = tf.reduce_max(gradient, axis=-1)
# Convert to NumPy & Remove batch axis
heatmap = gradient.numpy()[0]
# Intensity normalization to [0,1]
numer = heatmap - np.min(heatmap)
denom = (heatmap.max() - heatmap.min()) + eps
heatmap = numer / denom
# Return the resulting heatmap
return heatmap
|