def user_test_cam_ssd_fd(dev_idx, _user_id, test_loop):
"""User test cam yolo."""
image_source_h = 480
image_source_w = 640
app_id = constants_kl520.APP_FD_LM
image_size = image_source_w * image_source_h * 2
frames = []
# Setup video capture device.
capture = kdp_wrapper.setup_capture(0, image_source_w, image_source_h)
if capture is None:
return -1
# Start ISI mode.
if kdp_wrapper.start_isi(dev_idx, app_id, image_source_w, image_source_h):
return -1
# Fill up the image buffers.
ret, img_id_tx, img_left, buffer_depth = kdp_wrapper.fill_buffer(
dev_idx, capture, image_size, frames)
if ret:
return -1
# Send the rest and get result in loop, with 2 images alternatively
print("Companion image buffer depth = ", buffer_depth)
ret = kdp_wrapper.pipeline_inference(dev_idx, app_id, test_loop,
image_size, capture, img_id_tx,
img_left, buffer_depth, frames,
handle_result)
kdp_wrapper.kdp_exit_isi(dev_idx)
return ret
def handle_result(dev_idx, inf_res, r_size, frames):
"""Handle the detected results returned from the model.
Arguments:
dev_idx: Connected device ID. A host can connect several devices.
inf_res: Inference result data.
r_size: Inference data size.
frames: List of frames captured by the video capture instance.
"""
if r_size >= 4:
header_result = ctypes.cast(
ctypes.byref(inf_res),
ctypes.POINTER(constants_kl520.ObjectDetectionRes)).contents
box_result = ctypes.cast(
ctypes.byref(header_result.boxes),
ctypes.POINTER(constants_kl520.BoundingBox *
header_result.box_count)).contents
for box in box_result:
x1 = int(box.x1)
y1 = int(box.y1)
x2 = int(box.x2)
y2 = int(box.y2)
frames[0] = cv2.rectangle(frames[0], (x1, y1), (x2, y2),
(0, 0, 255), 3)
cv2.imshow('detection', frames[0])
del frames[0]
key = cv2.waitKey(1)
if key == ord('q'):
kdp_wrapper.kdp_exit_isi(dev_idx)
sys.exit()
return 0
def user_test_cam_sync_yolo(dev_idx, _user_id, test_loop):
"""User test cam yolo."""
image_source_w = 640
image_source_h = 480
image_size = image_source_w * image_source_h * 2
frames = []
# Setup video capture device.
capture = kdp_wrapper.setup_capture(0, image_source_w, image_source_h)
if capture is None:
return -1
# Start ISI mode.
if kdp_wrapper.start_isi(dev_idx, ISI_YOLO_ID, image_source_w,
image_source_h):
return -1
# Fill up the image buffers.
img_id_tx = 0
start_time = time.time()
while (img_id_tx != test_loop):
kdp_wrapper.sync_inference(dev_idx, ISI_YOLO_ID, image_size, capture,
img_id_tx, frames, handle_result)
img_id_tx += 1
end_time = time.time()
diff = end_time - start_time
estimate_runtime = float(diff / test_loop)
fps = float(1 / estimate_runtime)
print("Sync inference average estimate runtime is ", estimate_runtime)
print("Average FPS is ", fps)
kdp_wrapper.kdp_exit_isi(dev_idx)
return 0
def handle_result(dev_idx, raw_res, captured_frames):
"""Handle the raw results returned from the model.
Arguments:
dev_idx: Connected device ID. A host can connect several devices.
raw_res: Raw numpy data.
captured_frames: List of frames captured by the video capture instance.
"""
# the parameters for postprocess
anchor_path = './examples_kl520/yolo/models/anchors.txt'
class_path = './common/coco_name_lists'
model_input_shape = (224, 224)
score_thres = 0.2
nms_thres = 0.45
keep_aspect_ratio = True
image_source_h = 480
image_source_w = 640
det_res = yolo_postprocess_(raw_res, anchor_path, class_path, image_source_h, image_source_w, model_input_shape,
score_thres, nms_thres, keep_aspect_ratio)
#for multiple detection
for res in det_res:
x1 = int(res[0])
y1 = int(res[1])
x2 = int(res[2])
y2 = int(res[3])
class_num = res[5]
score = res[4]
# print(x1,x2,class_num,score)
if (class_num==0):
captured_frames[0] = cv2.rectangle(captured_frames[0], (x1, y1), (x2, y2), (0, 0, 255), 3)
# print("score of person: ", score)
else:
captured_frames[0] = cv2.rectangle(captured_frames[0], (x1, y1), (x2, y2), (255, 0, 0), 3)
# print("score of others: ", score)
cv2.imshow('detection', captured_frames[0])
del captured_frames[0]
key = cv2.waitKey(1)
if key == ord('q'):
kdp_wrapper.kdp_exit_isi(dev_idx)
sys.exit()
return
def user_test_cam_yolo(dev_idx, _user_id, test_loop):
"""User test cam yolo."""
is_raw_output = True
image_source_h = 480
image_source_w = 640
app_id = constants_kl520.APP_TINY_YOLO3
image_size = image_source_w * image_source_h * 2
frames = []
# Setup video capture device.
capture = kdp_wrapper.setup_capture(0, image_source_w, image_source_h)
if capture is None:
return -1
# Start ISI mode.
if kdp_wrapper.start_isi_parallel_ext(dev_idx, app_id, image_source_w, image_source_h, is_raw_output):
return -1
start_time = time.time()
# Fill up the image buffers.
ret, img_id_tx, img_left, buffer_depth = kdp_wrapper.fill_buffer(
dev_idx, capture, image_size, frames)
if ret:
return -1
# Send the rest and get result in loop, with 2 images alternatively
print("Companion image buffer depth = ", buffer_depth)
kdp_wrapper.pipeline_inference(
dev_idx, app_id, test_loop - buffer_depth, image_size,
capture, img_id_tx, img_left, buffer_depth, frames, handle_result, is_raw_output)
end_time = time.time()
diff = end_time - start_time
estimate_runtime = float(diff/test_loop)
fps = float(1/estimate_runtime)
print("Parallel inference average estimate runtime is ", estimate_runtime)
print("Average FPS is ", fps)
kdp_wrapper.kdp_exit_isi(dev_idx)
return 0