[WIP] adding in Security Semantic Segmentation interp script

[benhoff]

It should be explicitly noted that this script changes the preprocessing step of images.

This change set scales the images to preserve the aspect ratio (and then padding the other area with blank values). The previous preprocessing did not preserve the aspect ratio, and scaled the image to match the input parameters.

So the code in this change set is:

_, _, h, w = self._input_layout
scale = min(h / image.shape[0], w / image.shape[1])
in_frame = image if image.shape[:-1] == (h, w) else cv2.resize(image, None, fx=scale, fy=scale)

in_frame_size = in_frame.shape[:2]
in_frame = np.pad(in_frame, ((0, h - in_frame_size[0]), (0, w - in_frame_size[1]), (0, 0)), mode='constant', constant_values=0)

The previous code:

_, _, h, w = self._input_layout
in_frame = image if image.shape[:-1] == (h, w) else cv2.resize(image, (w, h))

I'm not sure if that would have an impact on previous models or not.

[benhoff]

I wouldn't recommend merging this, but it can be used to test #759

[benhoff]

Note to self, this looks wrong. im_w and im_h appear flipped. need to investigate.