def initialize_app(self):
app = Application(name="rm_isaac_bridge")
# app.load(filename="packages/navsim/apps/navsim_tcp.subgraph.json", prefix="simulation")
app.load(filename="packages/ros_bridge/apps/ros_to_perception.subgraph.json", prefix="ros_perception")
if hasattr(self, 'arm'):
app = self.arm.connect_app(app)
if hasattr(self, 'camera'):
app = self.camera.connect_app(app)
if hasattr(self, 'effector'):
app = self.effector.connect_app(app)
self._app = app
def main(args):
# Read CVAT XML file
cvat_xml_path = args.cvat_xml
if os.path.exists(cvat_xml_path):
tree = et.parse(cvat_xml_path)
else:
print("Please provide a valid XML file from CVAT.")
return
# Get image cask UUID that these labels are associated with
image_cask_uuid = cvat_xml_path.split('/')[-1].split('.')[0]
# Start application to record
app = Application()
# Add a dummy node to publish the constructed Detections2 messages from
app.add("node")
message_ledger = app.nodes["node"].components["MessageLedger"]
# Load record subgraph and configure
app.load("packages/cask/apps/record.subgraph.json", prefix="record")
record_interface = app.nodes["record.interface"].components["input"]
record_interface.config.base_directory = args.base_directory_gt
# Connect output of dummy node to recorder
app.connect(message_ledger, 'in', record_interface, 'bounding_boxes')
app.start()
# Loop through each image element in the XML tree
count = 0
for image in tree.findall("./image"):
# "Name" attribute corresponds to the image filepath that was input to the CVAT labeling
# tool. Convention is: <image_cask_uuid>/<channel>/<acqtime>.png
image_uuid, channel, png = image.attrib['name'].split('/')
# Check that the image_uuid corresponds to the one specified by the XML filename
if (image_uuid != image_cask_uuid): continue
# Extract the acquisition time
acqtime = int(png.lstrip('0').split('.')[0])
# S the detections of interest for this image
all_ground_truth_bboxes = image.findall("./box")
sliced_ground_truth_boxes = slice_detections(all_ground_truth_bboxes,
args.slice_mode)
num_sliced_ground_truth_boxes = len(sliced_ground_truth_boxes)
# Build Detections2Proto message
detections2 = Message.create_message_builder('Detections2Proto')
detections2.acqtime = acqtime
detections2.uuid = str(uuid.uuid1())
predictions = detections2.proto.init('predictions',
num_sliced_ground_truth_boxes)
bounding_boxes = detections2.proto.init('boundingBoxes',
num_sliced_ground_truth_boxes)
# Populate the Detections2Proto and PredictionProto messages per sliced boudning box
for i in range(num_sliced_ground_truth_boxes):
box = sliced_ground_truth_boxes[i]
row = {
a.attrib['name']: a.text
for a in box.findall("./attribute")
}
row.update(box.attrib)
prediction = predictions[i]
prediction.label = row['label']
prediction.confidence = 1.0
bbox = bounding_boxes[i]
bbox.min.y = float(row['xtl'])
bbox.min.x = float(row['ytl'])
bbox.max.y = float(row['xbr'])
bbox.max.x = float(row['ybr'])
# Publish the message to the node being recorded
app.publish('node', 'MessageLedger', 'in', detections2)
recv_msg = app.receive('node', 'MessageLedger', 'in')
count += 1
time.sleep(0.1) #sleep to make sure we don't lose any messages
app.stop()
print("Wrote " + str(count) + " messages")
# Write metadata to JSON data per output cask. The metadata servers to associate
# corresponding image and ground truth casks. As per RACI evaluation workflow
# and data management, image casks and ground truth casks are stored in separate
# directories.
if args.raci_metadata:
# Populate ground truth cask metadata
ground_truth_metadata_json = {}
ground_truth_metadata_json["Image_Cask_File"] = image_cask_uuid
ground_truth_metadata_json["Data_Source"] = "ground_truth"
# Write ground truth cask metadata
ground_truth_metadata_path = os.path.join(args.base_directory_gt,
app.uuid + "_md.json")
with open(ground_truth_metadata_path, 'w') as f:
json.dump(ground_truth_metadata_json, f, indent=2)