def process_image(data):
image = process_compressed_image(data,
sensor_stats={
'height': 256,
'width': 256,
'depth': 3
})
output_images = [PILImage.fromarray((image * 255).astype(np.uint8))]
o_t = TF.to_tensor(image) * 2 - 1
o_t = o_t.unsqueeze_(0)
task = 'segment_semantic'
pred = visualpriors.feature_readout(o_t, task, device='cpu')
labels = semantic_masks(pred)
output_images.append(PILImage.fromarray(labels))
widths, heights = zip(*(i.size for i in output_images))
total_width = sum(widths)
max_height = max(heights)
new_im = PILImage.new('RGB', (total_width, max_height))
x_offset = 0
for im in output_images:
new_im.paste(im, (x_offset, 0))
x_offset += im.size[0]
cv2.imshow('image', np.asarray(new_im))
cv2.waitKey(1)
def _process_compressed_image(self, msg: CompressedImage,
args: tuple) -> None:
sensor_topic, sensor_stats = args
image = process_compressed_image(msg, sensor_stats)
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
if sum(image.shape) < 3000:
image = cv2.resize(image, (600, 400), interpolation=cv2.INTER_AREA)
self._draw(image)
def _process_observation(self):
if self._view_msg is None:
return None
msg = copy.deepcopy(self._view_msg)
self._view_msg = None
image = process_image(msg,
self._observation_sensor_stats) if isinstance(
msg, Image) else process_compressed_image(
msg, self._observation_sensor_stats)
return cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
def _process_mask(self):
if self._mask_msg is None:
return None
msg = copy.deepcopy(self._mask_msg)
self._mask_msg = None
image = process_image(msg, self._mask_sensor_stats) if isinstance(
msg, Image) else process_compressed_image(msg,
self._mask_sensor_stats)
image -= image.min()
image /= image.max()
image *= 255
return cv2.applyColorMap(image.astype(np.uint8),
cv2.COLORMAP_AUTUMN) / 255.