d6fca70d938e502efff15f6b71bf38cac7fcabbe,src/attackers/deepfool.py,DeepFool,generate,#DeepFool#Any#,41

Before Change

        xi_op = tf.placeholder(dtype=tf.float32, shape=dims)

        // Using model predictions as ground truth to avoid label leaking
        preds_max = tf.reduce_max(self.model(xi_op), 1, keep_dims=True)
        y = tf.to_float(tf.equal(self.model(xi_op), preds_max))
        y = y / tf.reduce_sum(y, 1, keep_dims=True)

        // Compute loss
        loss = model_loss(y, self.model(xi_op), mean=False)
        grad_xi, = tf.gradients(loss, xi_op)

After Change

                nb_iter += 1

            progress_bar.update(current=j, values=[("perturbation", abs(np.average(r)))])

        return x_adv

Instances