Script to preprocess input image and get a feature map

pre_processing.py

def pre_process(img):
    """ Prepare image to be feed to a Keras convolutive model.
    
    img: 3-channel Numpy array image of size.
    Returns a resized and smoothed 250x70 RGB image. 
    """
    # Crop relevant area from input image
    img = crop(img,600,1080,0,1800)
    # Resize to model's input size (250,70)
    img = cv2.resize(img,(250,70), interpolation = cv2.INTER_LANCZOS4)#INTER_CUBIC)
    # Smooth image with mean filter
    kernel = np.ones((2,2),np.float32)/4
    img = cv2.filter2D(img,-1,kernel)

    return img

# Save input image ROI 
cropped_img = crop(imgs[0],600,1080,0,1800)

# Transform original image to model's input size (250, 70, 3)
dst = pre_process(imgs[0])

# Get desired feature map (2nd convolutive layer)
layer_name = model.layers[6].name

# Input image to Keras model. Don't forget the mini-batch dimension
activations = get_activations(model,np.expand_dims(dst, axis=0), 
                              print_shape_only=True, 
                              layer_name=layer_name)

# Normalize heatmap to pixels from (0 - 255)
target_map = norm_pixels(activations[0][0][:,:,1])