gradcam

This file includes the GradCAM class, which implements the GradCAM algorithm for visualizing and analyzing deep neural networks.

Initialization

The __init__ method initilizes the GradCAM object by setting the provided model and layer name. If a layer name in not provided, the last convolutional layer will be used.

def __init__(self, model=None, layer_name=""):
   print("init gradcam")
   print(model)
   self.model = model

   self.layer_name = layer_name
   if layer_name!="":
       self.layer_name = layer_name
       try:
           layer=self.model.get_layer(self.layer_name)
       except Exception as e:
           print(f"Error getting the layer: {e}", "reverting to last convolutional layer")
           self.layer_name = ""

   if self.layer_name=="" and self.model is not None:
       for layer in reversed(self.model.layers):
           if isinstance(layer, tf.keras.layers.Conv2D):
               self.layer_name = layer.name
               break


   if self.layer_name == "":
       print("Error finding the last convolutional layer")

   print("layer name:",self.layer_name)

   try:
       self.grad_model = tf.keras.models.Model([self.model.inputs], [self.model.get_layer(self.layer_name).output, self.model.output])
   except Exception as e:
       print(f"Error getting the model: {e}")
       self.grad_model = None

Parameters

model: An optional object representing a trained AI model.
layer_name: An optional string corresponding to a specific layer in the AI model.

get_heatmap

This method generates the GradCAM heatmap for the given image path.

def get_heatmap(self, path_to_image, pred_index=None):
    print("getting heatmap")
    print(self.grad_model)
    if self.grad_model is None:
        return None

    input_shape=self.model.input_shape[1:]
    print("input shape",input_shape)
    print("reading image:",path_to_image)
    image = cv2.imread(path_to_image, cv2.IMREAD_COLOR)
    #image = cv2.resize(image, input_shape[:2])
    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
    image = image.astype('float32')
    image = cv2.normalize(image, None, 0, 1, cv2.NORM_MINMAX)
    #image=image.transpose(1,0,2)
    tensor = tf.convert_to_tensor(image)
    tensor = np.expand_dims(tensor, axis=0)
    print("post tensor")
    with tf.GradientTape() as tape:
        last_conv_layer_output, preds = self.grad_model(tensor)
        if pred_index is None:
            pred_index = tf.argmax(preds[0])
        class_channel = preds[:, pred_index]

    grads = tape.gradient(class_channel, last_conv_layer_output)
    pooled_grads = tf.reduce_mean(grads, axis=(0, 1, 2))
    last_conv_layer_output = last_conv_layer_output[0]
    heatmap = last_conv_layer_output @ pooled_grads[..., tf.newaxis]
    heatmap = tf.squeeze(heatmap)
    heatmap = tf.maximum(heatmap, 0) / tf.math.reduce_max(heatmap)

    heatmap_rgb = cm.coolwarm(heatmap)[:, :, :3] * 255
    heatmap_rgb = heatmap_rgb.astype('uint8')
    heatmap = cv2.resize(heatmap_rgb, (tensor.shape[2], tensor.shape[1]))

    print(preds)
    print(pred_index)
    return heatmap, preds, pred_index

# add in a mysqldb connection to put the heatmap into the database

Parameters

path_to_image: A string representing the path to the image file.
pred_index: An optional integer representing the predicted class index.

Example Usage

if model_id!=-1:
       gradCAMheatmap, preds, top_class_index = self.get_heatmap(path_to_image)

insert_into_runtime

This method inserts the GradCAM heatmap and related information into the runtime database.

# Extended code available on Github
def insert_into_runtime(self, path_to_image, plot_type_id, model_id, runnum=0):

Parameters

path_to_image: A string representing the path to the image file.
plot_type_id: An integer representing the plot ID in the datbase.
model_id: An integer representing the model ID in the database. This is set to -1 if no model is available.
runnum: An optional integer representing the run number

Example Usage

if modelID==-1 :
        print("True")
        grad=GradCAM(None,-1)
        grad.insert_into_runtime(args["input"],-1,-1,runnum)