Fix bugs and Improve the approach of makemovie

[BillyCheung10botics]

1. fix the bugs originated from the latest commit b544241

The commit b544241 made a modification to the keras.py part on which the makemovie.py heavily depends. This results in errors of the code calling the keras part,

donkeycar/donkeycar/management/makemovie.py

Line 76 in 50f0831

self.do_salient = self.init_salient(self.keras_part.model)

donkeycar/donkeycar/management/makemovie.py

Lines 147 to 148 in 50f0831

	pred_img = pred_img.reshape((1,) + pred_img.shape)
	angle_binned, _ = self.keras_part.model.interpreter.predict(pred_img)

donkeycar/donkeycar/management/makemovie.py

Line 207 in 50f0831

expected = self.keras_part.model.inputs[0].shape[1:]

2. improve the approach to calculate saliency maps,

2.1 evaluate the effects of all output layers instead of one
2.2 correct the problematic math (blending the gradients of angle and throttle output layers together) from direct sum to square norm
2.3 calculate the saliency depending on the model type "linear" and "categorical"

3. modify the "draw_user_input", "draw_model_prediction", and "draw_steering_distribution" methods

I add an argument taking an image array to each method, such that the labels are drawn on this image array instead of the same image for inference. For instance, if we draw the user input green line on the input image, and then pass the same image for inference, that green line could affect the model prediction and give an incorrect prediction blue line.

These labels (yellow-purple cloud, blue line, green line, red line and grey lines) could affect the model prediction. It is better to seperate the image array for inference and the image array on which labels are drawn.

[sctse999]

@DocGarbanzo @naisy any further comment? If no I will merge this.

[DocGarbanzo]

Thanks for that contribution - Looks good to me. Just a minor comment.

[DocGarbanzo]

There is a member function get_input_shapes() in KerasPilot.