main.py 文件源码

def grad_cam(x, vgg, sess, predicted_class, layer_name, nb_classes):
    print("Setting gradients to 1 for target class and rest to 0")
    # Conv layer tensor [?,7,7,512]
    conv_layer = vgg.layers[layer_name]
    # [1000]-D tensor with target class index set to 1 and rest as 0
    one_hot = tf.sparse_to_dense(predicted_class, [nb_classes], 1.0)
    signal = tf.mul(vgg.layers['fc3'], one_hot)
    loss = tf.reduce_mean(signal)

    grads = tf.gradients(loss, conv_layer)[0]
    # Normalizing the gradients
    norm_grads = tf.div(grads, tf.sqrt(tf.reduce_mean(tf.square(grads))) + tf.constant(1e-5))

    output, grads_val = sess.run([conv_layer, norm_grads], feed_dict={vgg.imgs: x})
    output = output[0]           # [7,7,512]
    grads_val = grads_val[0]     # [7,7,512]

    weights = np.mean(grads_val, axis = (0, 1))             # [512]
    cam = np.ones(output.shape[0 : 2], dtype = np.float32)  # [7,7]

    # Taking a weighted average
    for i, w in enumerate(weights):
        cam += w * output[:, :, i]

    # Passing through ReLU
    cam = np.maximum(cam, 0)
    cam = cam / np.max(cam)
    cam = resize(cam, (224,224))

    # Converting grayscale to 3-D
    cam3 = np.expand_dims(cam, axis=2)
    cam3 = np.tile(cam3,[1,1,3])

    return cam3