def run_mask_detection(video_path,
detection_graph,
label_map,
categories,
category_index,
show_window,
visualize,
write_output,
ros_enabled,
usage_check,
graph_trace_enabled=False,
score_node=None,
expand_node=None):
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
from tf_object_detection.utils import ops as utils_ops
from PIL import Image
from tf_object_detection.utils import visualization_utils as vis_util
config = tf.ConfigProto(allow_soft_placement=True,
gpu_options=tf.GPUOptions(allow_growth=True))
labels_per_frame = []
boxes_per_frame = []
cpu_usage_dump = ""
mem_usage_dump = ""
time_usage_dump = ""
if ros_enabled:
from utils.ros_op import DetectionPublisher, CameraSubscriber
pub = DetectionPublisher()
sub = CameraSubscriber()
if graph_trace_enabled:
from tensorflow.python.client import timeline
if usage_check:
timer = Timer()
logger.info("Initial startup")
cpu_usage_dump, mem_usage_dump, time_usage_dump = show_usage(
cpu_usage_dump, mem_usage_dump, time_usage_dump, timer)
if ros_enabled:
if not sub.is_running():
return Exception("[ERROR: Camera Node not running]")
else:
vid = WebcamVideoStream(src=video_path).start()
r, c = vid.get_dimensions()
logger.debug("Frame width: {} height: {}".format(r, c))
if write_output:
trackedVideo = cv2.VideoWriter(
'output.avi', cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 20.0,
(c, r))
record = open("record.txt", "w")
count = 0
# Detection
with detection_graph.as_default():
with tf.Session(graph=detection_graph, config=config) as sess:
options = None
run_metadata = None
# Get handles to input and output tensors
ops = tf.get_default_graph().get_operations()
all_tensor_names = {
output.name
for op in ops for output in op.outputs
}
tensor_dict = {}
for key in [
'num_detections', 'detection_boxes', 'detection_scores',
'detection_classes', 'detection_masks'
]:
tensor_name = key + ':0'
if tensor_name in all_tensor_names:
tensor_dict[key] = tf.get_default_graph(
).get_tensor_by_name(tensor_name)
detection_boxes = tf.squeeze(tensor_dict['detection_boxes'], [0])
detection_masks = tf.squeeze(tensor_dict['detection_masks'], [0])
# Reframe is required to translate mask from box coordinates to image coordinates and fit the image size.
real_num_detection = tf.cast(tensor_dict['num_detections'][0],
tf.int32)
detection_boxes = tf.slice(detection_boxes, [0, 0],
[real_num_detection, -1])
detection_masks = tf.slice(detection_masks, [0, 0, 0],
[real_num_detection, -1, -1])
detection_masks_reframed = utils_ops.reframe_box_masks_to_image_masks(
detection_masks, detection_boxes, r, c)
detection_masks_reframed = tf.cast(
tf.greater(detection_masks_reframed, 0.5), tf.uint8)
# Follow the convention by adding back the batch dimension
tensor_dict['detection_masks'] = tf.expand_dims(
detection_masks_reframed, 0)
image_tensor = tf.get_default_graph().get_tensor_by_name(
'image_tensor:0')
# Using the split model hack
if score_node is not None and expand_node is not None:
score_out = detection_graph.get_tensor_by_name(
'Postprocessor/convert_scores:0')
expand_out = detection_graph.get_tensor_by_name(
'Postprocessor/ExpandDims_1:0')
score_in = detection_graph.get_tensor_by_name(
'Postprocessor/convert_scores_1:0')
expand_in = detection_graph.get_tensor_by_name(
'Postprocessor/ExpandDims_1_1:0')
if usage_check:
fps = FPS().start()
if graph_trace_enabled:
options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
# Read video frame by frame and perform inference
while (vid.is_running()):
try:
# the array based representation of the image will be used later in order to prepare the
# result image with boxes and labels on it.
logger.debug("Frame {}".format(count))
retval, curr_frame = vid.read()
if not retval:
logger.info("Video ending at frame {}".format(count))
break
if show_window:
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# Expand dimensions since the model expects images to have shape: [1, None, None, 3]
curr_frame_expanded = np.expand_dims(curr_frame, axis=0)
# Actual detection.
start = time.time()
if score_node is None and expand_node is None:
output_dict = sess.run(
tensor_dict,
feed_dict={image_tensor: curr_frame_expanded},
options=options,
run_metadata=run_metadata)
else:
raise Exception("Split model not supported for mask")
end = time.time()
boxes = output_dict['detection_boxes']
scores = output_dict['detection_scores']
classes = output_dict['detection_classes']
# all outputs are float32 numpy arrays, so convert types as appropriate
output_dict['num_detections'] = int(
output_dict['num_detections'][0])
output_dict['detection_classes'] = output_dict[
'detection_classes'][0].astype(np.uint8)
output_dict['detection_boxes'] = output_dict[
'detection_boxes'][0]
output_dict['detection_scores'] = output_dict[
'detection_scores'][0]
output_dict['detection_masks'] = output_dict[
'detection_masks'][0]
logger.info(output_dict['detection_masks'].shape)
if usage_check:
fps.update()
logger.info("Session run time: {:.4f}".format(end -
start))
logger.info("Frame {}".format(count))
cpu_usage_dump, mem_usage_dump, time_usage_dump = show_usage(
cpu_usage_dump, mem_usage_dump, time_usage_dump,
timer)
if graph_trace_enabled:
fetched_timeline = timeline.Timeline(
run_metadata.step_stats)
chrome_trace = fetched_timeline.generate_chrome_trace_format(
)
with open('graph_timeline.json', 'w') as f:
f.write(chrome_trace)
(r, c, _) = curr_frame.shape
logger.debug("image height:{}, width:{}".format(r, c))
# get boxes that pass the min requirements and their pixel coordinates
filtered_boxes = parse_tf_output(curr_frame.shape, boxes,
scores, classes)
if ros_enabled:
# TODO: Send the detected info to other systems every frame
logger.info("Publishing bboxes")
logger.info("".join([str(i) for i in filtered_boxes]))
pub.send_boxes(filtered_boxes)
if write_output:
record.write(str(count) + "\n")
for i in range(len(filtered_boxes)):
record.write("{}\n".format(str(filtered_boxes[i])))
# Visualization of the results of a detection.
if visualize:
# drawn_img = overlay(curr_frame, category_index, filtered_boxes)
vis_util.visualize_boxes_and_labels_on_image_array(
curr_frame,
output_dict['detection_boxes'],
output_dict['detection_classes'],
output_dict['detection_scores'],
category_index,
instance_masks=output_dict.get('detection_masks'),
use_normalized_coordinates=True,
line_thickness=8)
if show_window:
window_name = "stream"
cv2.namedWindow(window_name, cv2.WINDOW_NORMAL)
cv2.imshow(window_name, curr_frame)
if write_output:
trackedVideo.write(curr_frame)
else:
logger.info("".join([str(i) for i in filtered_boxes]))
count += 1
# Quick benchmarking
if timer.get_elapsed_time() >= 60:
break
except KeyboardInterrupt:
logger.info("Ctrl + C Pressed. Attempting graceful exit")
break
if usage_check:
fps.stop()
logger.info("[USAGE] elasped time: {:.2f}".format(fps.elapsed()))
logger.info("[USAGE] approx. FPS: {:.2f}".format(fps.fps()))
logger.info("[USAGE] inferenced frames: {}".format(fps.get_frames()))
logger.info("[USAGE] raw frames: {}".format(vid.get_raw_frames()))
logger.info("[USAGE] Total Time elapsed: {:.2f} seconds".format(
timer.get_elapsed_time()))
with open("cpu_usage.txt", "w") as c:
c.write(cpu_usage_dump)
with open("mem_usage.txt", "w") as m:
m.write(mem_usage_dump)
with open("time_usage.txt", "w") as t:
t.write(time_usage_dump)
vid.stop()
logger.debug("Result: {} frames".format(count))
if visualize:
cv2.destroyAllWindows()
if write_output:
record.close()
trackedVideo.release()
return labels_per_frame, boxes_per_frame
def run_segmentation(video_path,
detection_graph,
label_map,
categories,
category_index,
show_window,
visualize,
write_output,
ros_enabled,
usage_check,
graph_trace_enabled=False,
score_node=None,
expand_node=None):
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
from tf_object_detection.utils import ops as utils_ops
from PIL import Image
from tf_object_detection.utils import visualization_utils as vis_util
config = tf.ConfigProto(allow_soft_placement=True,
gpu_options=tf.GPUOptions(allow_growth=True))
labels_per_frame = []
boxes_per_frame = []
cpu_usage_dump = ""
mem_usage_dump = ""
time_usage_dump = ""
if ros_enabled:
from utils.ros_op import DetectionPublisher, CameraSubscriber
pub = DetectionPublisher()
sub = CameraSubscriber()
if graph_trace_enabled:
from tensorflow.python.client import timeline
if usage_check:
timer = Timer()
logger.info("Initial startup")
cpu_usage_dump, mem_usage_dump, time_usage_dump = show_usage(
cpu_usage_dump, mem_usage_dump, time_usage_dump, timer)
if ros_enabled:
if not sub.is_running():
return Exception("[ERROR: Camera Node not running]")
else:
vid = WebcamVideoStream(src=video_path).start()
r, c = vid.get_dimensions()
logger.debug("Frame width: {} height: {}".format(r, c))
if write_output:
trackedVideo = cv2.VideoWriter(
'output.avi', cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 20.0,
(c, r))
record = open("record.txt", "w")
count = 0
# Detection
with detection_graph.as_default():
with tf.Session(graph=detection_graph, config=config) as sess:
options = None
run_metadata = None
# Get handles to input and output tensors
ops = tf.get_default_graph().get_operations()
all_tensor_names = {
output.name
for op in ops for output in op.outputs
}
seg_tensor = "SemanticPredictions:0"
image_tensor = tf.get_default_graph().get_tensor_by_name(
'ImageTensor:0')
if usage_check:
fps = FPS().start()
if graph_trace_enabled:
options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
# Read video frame by frame and perform inference
while (vid.is_running()):
try:
# the array based representation of the image will be used later in order to prepare the
# result image with boxes and labels on it.
logger.debug("Frame {}".format(count))
retval, curr_frame = vid.read()
if not retval:
logger.info("Video ending at frame {}".format(count))
break
if show_window:
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# curr_frame = cv2.cvtColor(curr_frame, cv2.COLOR_BGR2RGB)
# Expand dimensions since the model expects images to have shape: [1, None, None, 3]
curr_frame_expanded = np.expand_dims(curr_frame, axis=0)
# Actual detection.
start = time.time()
if score_node is None and expand_node is None:
output_dict = sess.run(
seg_tensor,
feed_dict={image_tensor: curr_frame_expanded},
options=options,
run_metadata=run_metadata)
else:
raise Exception(
"Split model not supported for segmentation")
end = time.time()
if usage_check:
fps.update()
logger.info("Session run time: {:.4f}".format(end -
start))
logger.info("Frame {}".format(count))
cpu_usage_dump, mem_usage_dump, time_usage_dump = show_usage(
cpu_usage_dump, mem_usage_dump, time_usage_dump,
timer)
if graph_trace_enabled:
fetched_timeline = timeline.Timeline(
run_metadata.step_stats)
chrome_trace = fetched_timeline.generate_chrome_trace_format(
)
with open('graph_timeline.json', 'w') as f:
f.write(chrome_trace)
(r, c, _) = curr_frame.shape
logger.debug("image height:{}, width:{}".format(r, c))
if ros_enabled:
# TODO: Send the detected info to other systems every frame
logger.info("Publishing bboxes")
logger.info("".join([str(i) for i in filtered_boxes]))
pub.send_boxes(filtered_boxes)
if write_output:
record.write(str(count) + "\n")
for i in range(len(filtered_boxes)):
record.write("{}\n".format(str(filtered_boxes[i])))
# Visualization of the results of a detection.
if visualize:
logger.warning("visualize not implmented!")
else:
logger.info(output_dict.shape)
count += 1
# Quick benchmarking
if timer.get_elapsed_time() >= 60:
break
except KeyboardInterrupt:
logger.info("Ctrl + C Pressed. Attempting graceful exit")
break
if usage_check:
fps.stop()
logger.info("[USAGE] elasped time: {:.2f}".format(fps.elapsed()))
logger.info("[USAGE] approx. FPS: {:.2f}".format(fps.fps()))
logger.info("[USAGE] inferenced frames: {}".format(fps.get_frames()))
logger.info("[USAGE] raw frames: {}".format(vid.get_raw_frames()))
logger.info("[USAGE] Total Time elapsed: {:.2f} seconds".format(
timer.get_elapsed_time()))
with open("cpu_usage.txt", "w") as c:
c.write(cpu_usage_dump)
with open("mem_usage.txt", "w") as m:
m.write(mem_usage_dump)
with open("time_usage.txt", "w") as t:
t.write(time_usage_dump)
vid.stop()
logger.debug("Result: {} frames".format(count))
if visualize:
cv2.destroyAllWindows()
if write_output:
record.close()
trackedVideo.release()
return labels_per_frame, boxes_per_frame
def run_detection(video_path,
detection_graph,
label_map,
categories,
category_index,
show_window,
visualize,
write_output,
ros_enabled,
usage_check,
graph_trace_enabled=False,
score_node=None,
expand_node=None):
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
config = tf.ConfigProto(allow_soft_placement=True,
gpu_options=tf.GPUOptions(allow_growth=True))
labels_per_frame = []
boxes_per_frame = []
cpu_usage_dump = ""
mem_usage_dump = ""
time_usage_dump = ""
if ros_enabled:
from utils.ros_op import DetectionPublisher, CameraSubscriber
pub = DetectionPublisher()
sub = CameraSubscriber()
if graph_trace_enabled:
from tensorflow.python.client import timeline
if usage_check:
timer = Timer()
logger.info("Initial startup")
cpu_usage_dump, mem_usage_dump, time_usage_dump = show_usage(
cpu_usage_dump, mem_usage_dump, time_usage_dump, timer)
if ros_enabled:
if not sub.is_running():
return Exception("[ERROR: Camera Node not running]")
else:
vid = WebcamVideoStream(src=video_path).start()
r, c = vid.get_dimensions()
logger.debug("Frame width: {} height: {}".format(r, c))
if write_output:
trackedVideo = cv2.VideoWriter(
'output.avi', cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 20.0,
(c, r))
record = open("record.txt", "w")
count = 0
# Detection
with detection_graph.as_default():
with tf.Session(graph=detection_graph, config=config) as sess:
options = None
run_metadata = None
# Definite input and output Tensors for detection_graph
image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
# Each box represents a part of the image where a particular object was detected.
detection_boxes = detection_graph.get_tensor_by_name(
'detection_boxes:0')
# Each score represent how level of confidence for each of the objects.
# Score is shown on the result image, together with the class label.
detection_scores = detection_graph.get_tensor_by_name(
'detection_scores:0')
detection_classes = detection_graph.get_tensor_by_name(
'detection_classes:0')
# Using the split model hack
if score_node is not None and expand_node is not None:
score_out = detection_graph.get_tensor_by_name(
'Postprocessor/convert_scores:0')
expand_out = detection_graph.get_tensor_by_name(
'Postprocessor/ExpandDims_1:0')
score_in = detection_graph.get_tensor_by_name(
'Postprocessor/convert_scores_1:0')
expand_in = detection_graph.get_tensor_by_name(
'Postprocessor/ExpandDims_1_1:0')
if usage_check:
fps = FPS().start()
if graph_trace_enabled:
options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
run_metadata = tf.RunMetadata()
# Read video frame by frame and perform inference
while (vid.is_running()):
try:
# the array based representation of the image will be used later in order to prepare the
# result image with boxes and labels on it.
logger.debug("Frame {}".format(count))
retval, curr_frame = vid.read()
if not retval:
logger.info("Video ending at frame {}".format(count))
break
if show_window:
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# Expand dimensions since the model expects images to have shape: [1, None, None, 3]
curr_frame_expanded = np.expand_dims(curr_frame, axis=0)
curr_frame_expanded = np.int8(curr_frame_expanded)
# Actual detection.
start = time.time()
if score_node is None and expand_node is None:
(boxes, scores, classes) = sess.run(
[
detection_boxes, detection_scores,
detection_classes
],
feed_dict={image_tensor: curr_frame_expanded},
options=options,
run_metadata=run_metadata)
if graph_trace_enabled:
write_trace(run_metadata, timeline,
"graph_timeline_nosplit.json")
else:
# Split Detection in two sessions.
(score, expand) = sess.run(
[score_out, expand_out],
feed_dict={image_tensor: curr_frame_expanded},
options=options,
run_metadata=run_metadata)
if graph_trace_enabled:
write_trace(run_metadata, timeline,
"graph_timeline_conv.json")
(boxes, scores,
classes) = sess.run([
detection_boxes, detection_scores,
detection_classes
],
feed_dict={
score_in: score,
expand_in: expand
},
options=options,
run_metadata=run_metadata)
if graph_trace_enabled:
write_trace(run_metadata, timeline,
"graph_timeline_nms.json")
end = time.time()
if usage_check:
fps.update()
logger.info("Session run time: {:.4f}".format(end -
start))
logger.info("Frame {}".format(count))
cpu_usage_dump, mem_usage_dump, time_usage_dump = show_usage(
cpu_usage_dump, mem_usage_dump, time_usage_dump,
timer)
(r, c, _) = curr_frame.shape
logger.debug("image height:{}, width:{}".format(r, c))
# get boxes that pass the min requirements and their pixel coordinates
filtered_boxes = parse_tf_output(curr_frame.shape, boxes,
scores, classes)
if ros_enabled:
# TODO: Send the detected info to other systems every frame
logger.info("Publishing bboxes")
logger.info("".join([str(i) for i in filtered_boxes]))
pub.send_boxes(filtered_boxes)
if write_output:
record.write(str(count) + "\n")
for i in range(len(filtered_boxes)):
record.write("{}\n".format(str(filtered_boxes[i])))
# Visualization of the results of a detection.
if visualize:
drawn_img = overlay(curr_frame, category_index,
filtered_boxes)
if show_window:
window_name = "stream"
cv2.namedWindow(window_name, cv2.WINDOW_NORMAL)
cv2.imshow(window_name, drawn_img)
if write_output:
trackedVideo.write(drawn_img)
else:
logger.info("".join([str(i) for i in filtered_boxes]))
count += 1
# Quick benchmarking
if timer.get_elapsed_time() >= 60:
break
except KeyboardInterrupt:
logger.info("Ctrl + C Pressed. Attempting graceful exit")
break
if usage_check:
fps.stop()
logger.info("[USAGE] elasped time: {:.2f}".format(fps.elapsed()))
logger.info("[USAGE] approx. FPS: {:.2f}".format(fps.fps()))
logger.info("[USAGE] inferenced frames: {}".format(fps.get_frames()))
logger.info("[USAGE] raw frames: {}".format(vid.get_raw_frames()))
logger.info("[USAGE] Total Time elapsed: {:.2f} seconds".format(
timer.get_elapsed_time()))
with open("cpu_usage.txt", "w") as c:
c.write(cpu_usage_dump)
with open("mem_usage.txt", "w") as m:
m.write(mem_usage_dump)
with open("time_usage.txt", "w") as t:
t.write(time_usage_dump)
vid.stop()
logger.debug("Result: {} frames".format(count))
if visualize:
cv2.destroyAllWindows()
if write_output:
record.close()
trackedVideo.release()
return labels_per_frame, boxes_per_frame
# # draw the bounding box of the face along with the associated
# # probability
# text = "{:.2f}%".format(confidence * 100)
# y = startY - 10 if startY - 10 > 10 else startY + 10
# cv2.rectangle(frame, (startX, startY), (endX, endY),
# (0, 0, 255), 2)
# cv2.putText(frame, text, (startX, y),
# cv2.FONT_HERSHEY_SIMPLEX, 0.45, (0, 0, 255), 2)
# show the output frame
cv2.imshow("Didux.io", frame)
cv2.setWindowProperty('Didux.io', cv2.WND_PROP_ASPECT_RATIO,
cv2.WINDOW_FREERATIO)
cv2.setWindowProperty('Didux.io', cv2.WND_PROP_FULLSCREEN,
cv2.WINDOW_FULLSCREEN)
fps.update()
if cv2.waitKey(30) & 0xFF == ord('q'):
break
if fps._numFrames < args["num_frames"]:
fps.update()
if fps._numFrames == args["num_frames"]:
# stop the timer and display FPS information
fps.stop()
print("[INFO] elasped time: {:.2f}".format(fps.elapsed()))
print("[INFO] approx. FPS: {:.2f}".format(fps.fps()))
fps.start()
cv2.destroyAllWindows()
cv2.waitKey(1)
vs.stop()
def run_detection(video_path,
model_path,
model_name,
weights_path,
classes,
show_window=True,
visualize=True,
write_output=False,
is_cpu=False,
ros_enabled=False,
usage_check=False):
# Calling the class of the model
spec = importlib.util.spec_from_file_location(model_name, model_path)
mod = importlib.util.module_from_spec(spec)
spec.loader.exec_module(mod)
Net = getattr(mod, model_name)
model = Net(classes)
model = torch.nn.DataParallel(model)
model = load_my_state_dict(model, torch.load(weights_path))
model.eval()
cpu_usage_dump = ""
mem_usage_dump = ""
time_usage_dump = ""
if not is_cpu:
model = model.cuda()
else:
return Exception("[ERROR: CPU mode not implemented]")
if usage_check:
timer = Timer()
logger.info("Initial startup")
cpu_usage_dump, mem_usage_dump, time_usage_dump = show_usage(
cpu_usage_dump, mem_usage_dump, time_usage_dump, timer)
if ros_enabled:
if not sub.is_running():
return Exception("[ERROR: Camera Node not running]")
else:
vid = WebcamVideoStream(src=video_path).start()
r, c = vid.get_dimensions()
logger.info("Video frame width: {} height: {}".format(r, c))
if write_output:
trackedVideo = cv2.VideoWriter(
'output.avi', cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 20.0,
(c, r))
record = open("record.txt", "w")
count = 0
if usage_check:
fps = FPS().start()
# Read video frame by frame and perform inference
while (vid.is_running()):
try:
# the array based representation of the image will be used later in order to prepare the
# result image with boxes and labels on it.
logger.debug("Frame {}".format(count))
retval, curr_frame = vid.read()
if not retval:
logger.info("Video ending at frame {}".format(count))
break
if show_window:
if cv2.waitKey(1) & 0xFF == ord('q'):
break
start = time.time()
# Convert numpy img to PyTorch Tensor, then expand dimension for model
convert = Compose([ToTensor()])
img_tensor = convert(curr_frame)
img_tensor = img_tensor.unsqueeze(0)
con = time.time()
logger.debug("img conversion time: {:.4f}".format(con - start))
if (not is_cpu):
image = img_tensor.cuda()
inputs = Variable(image, volatile=True)
outputs = model(inputs)
out = time.time()
logger.debug("output time: {:.4f}".format(out - con))
# Visualization only works on cpu tensor
if visualize:
label = outputs[0].max(
0)[1].byte().cpu().data # Mask to be published
else:
label = outputs[0].max(
0)[1].byte().data # Mask to be published
l = time.time()
logger.debug("labeling time: {:.4f}".format(l - out))
end = time.time()
if usage_check:
fps.update()
logger.info("Session run time: {:.4f}".format(end - start))
logger.info("Frame {}".format(count))
cpu_usage_dump, mem_usage_dump, time_usage_dump = show_usage(
cpu_usage_dump, mem_usage_dump, time_usage_dump, timer)
# TODO: Publish Segmentation
if ros_enabled:
logger.info("Publishing segmengatation via ROS")
else:
logger.info("Publishing segmentation via custom module")
# Visualization of the results of a detection.
if visualize:
# Visualizes based off of cityscape classes; this step takes a ton of time!
label_color = Colorize()(label.unsqueeze(0))
label_color = np.moveaxis(label_color.numpy(), 0, -1)
label_color = label_color[..., ::-1]
vis = time.time()
logger.debug("visualization time: {:.4f}".format(vis - end))
if show_window:
window_name = "stream"
cv2.namedWindow(window_name, cv2.WINDOW_NORMAL)
cv2.imshow(window_name, label_color)
if write_output:
trackedVideo.write(label_color)
count += 1
# # Quick benchmarking
# if timer.get_elapsed_time() >= 60:
# break
except KeyboardInterrupt:
logger.info("Ctrl + C Pressed. Attempting graceful exit")
break
if usage_check:
fps.stop()
logger.info("[USAGE] elasped time: {:.2f}".format(fps.elapsed()))
logger.info("[USAGE] approx. FPS: {:.2f}".format(fps.fps()))
logger.info("[USAGE] inferenced frames: {}".format(fps.get_frames()))
logger.info("[USAGE] raw frames: {}".format(vid.get_raw_frames()))
logger.info("[USAGE] Total Time elapsed: {:.2f} seconds".format(
timer.get_elapsed_time()))
with open("cpu_usage.txt", "w") as c:
c.write(cpu_usage_dump)
with open("mem_usage.txt", "w") as m:
m.write(mem_usage_dump)
with open("time_usage.txt", "w") as t:
t.write(time_usage_dump)
vid.stop()
logger.debug("Result: {} frames".format(count))
if visualize:
cv2.destroyAllWindows()
if write_output:
record.close()
trackedVideo.release()
def detect(self):
#TODO: make a generalized detection workflow
labels_per_frame = []
boxes_per_frame = []
if self.benchmark:
from benchmark.usage import Timer, UsageTracker
self.logger.info("Initial startup")
timer = Timer()
usage = UsageTracker(timer)
usage.get_usage()
self.logger.debug("Frame width: {} height: {}".format(
self.width, self.height))
if self.write_output:
self.trackedVideo = cv2.VideoWriter(
'output.avi', cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 20.0,
(self.width, self.height))
self.record = open("record.txt", "w")
count = 0
if self.benchmark:
fps = FPS().start()
while self.feed.is_running:
try:
# the array based representation of the image will be used later in order to prepare the
# result image with boxes and labels on it.
self.logger.debug("Frame {}".format(count))
retval, curr_frame = self.feed.read()
if not retval:
self.logger.info("Video ending at frame {}".format(count))
break
if self.show_stream:
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# Actual detection.
start = time.time()
output = self.model.inference(curr_frame)
end = time.time()
if self.task == "segmentation" and self.library == "pytorch":
mask = output.data
if self.benchmark:
fps.update()
self.logger.info("Session run time: {:.4f}".format(end -
start))
self.logger.info("Frame {}".format(count))
usage.get_usage()
# TODO: Publish Output
if self.ros_enabled:
self.logger.info("Publishing via ROS")
else:
self.logger.info("Publishing via custom module")
if self.show_stream:
#TODO: set which type of visualization to use based on task
if self.task == "segmentation" and self.library == "pytorch":
vis_output = output.cpu().data
self._visualize(self.task, vis_output)
count += 1
# Quick benchmarking
if timer.get_elapsed_time() >= 60:
break
except KeyboardInterrupt:
self.logger.info("Ctrl + C Pressed. Attempting graceful exit")
break
if self.benchmark:
fps.stop()
self.logger.info("[USAGE] elasped time: {:.2f}".format(
fps.elapsed()))
self.logger.info("[USAGE] approx. FPS: {:.2f}".format(fps.fps()))
self.logger.info("[USAGE] inferenced frames: {}".format(
fps.get_frames()))
self.logger.info("[USAGE] raw frames: {}".format(
self.feed.get_raw_frames()))
self.logger.info(
"[USAGE] Total Time elapsed: {:.2f} seconds".format(
timer.get_elapsed_time()))
usage.dump_usage()
self.feed.stop()
self.logger.debug("Result: {} frames".format(count))
if self.show_stream:
cv2.destroyAllWindows()
if self.write_output:
self.record.close()
self.trackedVideo.release()