def main():
args = parse_args()
# context list
if args.use_gpu:
ctx = mx.gpu(0)
else:
ctx = mx.cpu()
if args.version == 'v1':
from config_farm import configuration_10_320_20L_5scales_v1 as cfg
symbol_file_path = '../symbol_farm/symbol_10_560_25L_8scales_v1_deploy.json'
model_file_path = '../saved_model/configuration_10_560_25L_8scales_v1/train_10_560_25L_8scales_v1_iter_1400000.params'
elif args.version == 'v2':
from config_farm import configuration_10_320_20L_5scales_v2 as cfg
symbol_file_path = '../symbol_farm/symbol_10_320_20L_5scales_v2_deploy.json'
model_file_path = '../saved_model/configuration_10_320_20L_5scales_v2/train_10_320_20L_5scales_v2_iter_1800000.params'
else:
raise TypeError('Unsupported LFFD Version.')
face_predictor = predict.Predict(
mxnet=mx,
symbol_file_path=symbol_file_path,
model_file_path=model_file_path,
ctx=ctx,
receptive_field_list=cfg.param_receptive_field_list,
receptive_field_stride=cfg.param_receptive_field_stride,
bbox_small_list=cfg.param_bbox_small_list,
bbox_large_list=cfg.param_bbox_large_list,
receptive_field_center_start=cfg.param_receptive_field_center_start,
num_output_scales=cfg.param_num_output_scales)
if args.mode == 'image':
data_folder = args.data
file_name_list = [file_name for file_name in os.listdir(data_folder) \
if file_name.lower().endswith('jpg')]
for file_name in file_name_list:
im = cv2.imread(os.path.join(data_folder, file_name))
bboxes, infer_time = face_predictor.predict(im, resize_scale=1, score_threshold=0.6, top_k=10000, \
NMS_threshold=0.4, NMS_flag=True, skip_scale_branch_list=[])
for bbox in bboxes:
cv2.rectangle(im, (bbox[0], bbox[1]), (bbox[2], bbox[3]),
(0, 255, 0), 2)
# if max(im.shape[:2]) > 1600:
# scale = 1600/max(im.shape[:2])
# im = cv2.resize(im, (0, 0), fx=scale, fy=scale)
cv2.imshow('im', im)
cv2.waitKey(5000)
cv2.imwrite(
os.path.join(data_folder,
file_name.replace('.jpg', '_result.png')), im)
elif args.mode == 'video':
# win_name = 'LFFD DEMO'
# cv2.namedWindow(win_name, cv2.WINDOW_NORMAL)
data_folder = args.data
file_name_list = [file_name for file_name in os.listdir(data_folder) \
if file_name.lower().endswith('mp4')]
for file_name in file_name_list:
out_file = os.path.join(
data_folder, file_name.replace('.mp4', '_v2_gpu_result.avi'))
cap = cv2.VideoCapture(os.path.join(data_folder, file_name))
vid_writer = cv2.VideoWriter(out_file, \
cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 60, \
(int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)), \
int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))))
while cv2.waitKey(1) < 0:
ret, frame = cap.read()
if ret:
h, w, c = frame.shape
if not ret:
print("Done processing of %s" % file_name)
print("Output file is stored as %s" % out_file)
cv2.waitKey(3000)
break
tic = time.time()
bboxes, infer_time = face_predictor.predict(frame, resize_scale=1, score_threshold=0.6, top_k=10000, \
NMS_threshold=0.4, NMS_flag=True, skip_scale_branch_list=[])
toc = time.time()
detect_time = (toc - tic) * 1000
face_num = 0
for bbox in bboxes:
face_num += 1
cv2.rectangle(frame, (bbox[0], bbox[1]),
(bbox[2], bbox[3]), (0, 255, 0), 2)
computing_platform = 'Computing platform: NVIDIA GPU FP32'
cv2.putText(frame, computing_platform, (5, 30),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
input_resolution = 'Network input resolution: %sx%s' % (w, h)
cv2.putText(frame, input_resolution, (5, 65),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
infer_time_info = 'Inference time: %.2f ms' % (infer_time)
cv2.putText(frame, infer_time_info, (5, 100),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
infer_speed = 'Inference speed: %.2f FPS' % (1000 / infer_time)
cv2.putText(frame, infer_speed, (5, 135),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
face_num_info = 'Face num: %d' % (face_num)
cv2.putText(frame, face_num_info, (5, 170),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
vid_writer.write(frame.astype(np.uint8))
# cv2.imshow(win_name, frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
else:
raise TypeError('Unsupported File Format.')
def main():
args = parse_args()
# context list
if args.use_gpu:
ctx = mx.gpu(0)
else:
ctx = mx.cpu()
if args.version == 'v1':
from config_farm import configuration_10_320_20L_5scales_v1 as cfg
symbol_file_path = '../symbol_farm/symbol_10_560_25L_8scales_v1_deploy.json'
model_file_path = '../saved_model/configuration_10_560_25L_8scales_v1/train_10_560_25L_8scales_v1_iter_1400000.params'
elif args.version == 'v2':
from config_farm import configuration_10_320_20L_5scales_v2 as cfg
start = time.time()
symbol_file_path = '../symbol_farm/symbol_10_320_20L_5scales_v2_deploy.json'
end = time.time()
print("symbol_file_path time: {:.3f}s".format(end - start))
start = time.time()
model_file_path = '../saved_model/configuration_10_320_20L_5scales_v2/train_10_320_20L_5scales_v2_iter_1000000.params'
end = time.time()
print("model_file_path time: {:.3f}s".format(end - start))
else:
raise TypeError('Unsupported LFFD Version.')
start = time.time()
face_predictor = predict.Predict(
mxnet=mx,
symbol_file_path=symbol_file_path,
model_file_path=model_file_path,
ctx=ctx,
receptive_field_list=cfg.param_receptive_field_list,
receptive_field_stride=cfg.param_receptive_field_stride,
bbox_small_list=cfg.param_bbox_small_list,
bbox_large_list=cfg.param_bbox_large_list,
receptive_field_center_start=cfg.param_receptive_field_center_start,
num_output_scales=cfg.param_num_output_scales)
end = time.time()
print("Predict Constructor time: {:.3f}s".format(end - start))
if args.mode == 'image':
data_folder = args.data
start = time.time()
file_name_list = [file_name for file_name in os.listdir(data_folder) \
if file_name.lower().endswith('jpg')]
end = time.time()
print("Get files time: {:.3f}s".format(end - start))
for file_name in file_name_list:
im = cv2.imread(os.path.join(data_folder, file_name))
start = time.time()
bboxes, infer_time = face_predictor.predict(im, resize_scale=1, score_threshold=0.6, top_k=10000, \
NMS_threshold=0.4, NMS_flag=True, skip_scale_branch_list=[])
end = time.time()
print("Prediction time: {:.3f}s\n".format(end - start))
for bbox in bboxes:
cv2.rectangle(im, (bbox[0], bbox[1]), (bbox[2], bbox[3]),
(0, 255, 0), 2)
# if max(im.shape[:2]) > 1600:
# scale = 1600/max(im.shape[:2])
# im = cv2.resize(im, (0, 0), fx=scale, fy=scale)
cv2.imshow('im', im)
cv2.waitKey(50)
cv2.imwrite(
os.path.join(data_folder,
file_name.replace('.jpg', '_result.png')), im)
elif args.mode == 'video':
timeRecSave = []
frame_count = 0
# win_name = 'LFFD DEMO'
# cv2.namedWindow(win_name, cv2.WINDOW_NORMAL)
data_folder = args.data
file_name_list = [file_name for file_name in os.listdir(data_folder) \
if file_name.lower().endswith('mp4')]
for file_name in file_name_list:
out_file = os.path.join(
data_folder, file_name.replace('.mp4', '_v2_gpu_result.avi'))
cap = cv2.VideoCapture(os.path.join(data_folder, file_name))
vid_writer = cv2.VideoWriter(out_file, \
cv2.VideoWriter_fourcc('M', 'J', 'P', 'G'), 60, \
(int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)), \
int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT))))
while cv2.waitKey(1) < 0:
ret, frame = cap.read()
if ret:
h, w, c = frame.shape
frame_count += 1
if not ret:
print("Done processing of %s" % file_name)
print("Output file is stored as %s" % out_file)
cv2.waitKey(3000)
break
tic = time.time()
bboxes, infer_time = face_predictor.predict(frame, resize_scale=1, score_threshold=0.6, top_k=10000, \
NMS_threshold=0.4, NMS_flag=True, skip_scale_branch_list=[])
toc = time.time()
detect_time = (toc - tic) * 1000
timeRecSave.append(detect_time)
face_num = 0
for bbox in bboxes:
face_num += 1
cv2.rectangle(frame, (bbox[0], bbox[1]),
(bbox[2], bbox[3]), (0, 255, 0), 2)
computing_platform = 'Computing platform: NVIDIA GPU FP32'
cv2.putText(frame, computing_platform, (5, 30),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
input_resolution = 'Network input resolution: %sx%s' % (w, h)
cv2.putText(frame, input_resolution, (5, 65),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
infer_time_info = 'Inference time: %.2f ms' % (infer_time)
cv2.putText(frame, infer_time_info, (5, 100),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
infer_speed = 'Inference speed: %.2f FPS' % (1000 / infer_time)
cv2.putText(frame, infer_speed, (5, 135),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
face_num_info = 'Face num: %d' % (face_num)
cv2.putText(frame, face_num_info, (5, 170),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 3)
vid_writer.write(frame.astype(np.uint8))
# cv2.imshow(win_name, frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
timeRecSave.pop(0)
timeRecMean = np.mean(timeRecSave)
# %% Visualization of results
plt.figure(figsize=(8, 5))
plt.plot(timeRecSave, label='detection time')
plt.ylim([0, 82])
plt.xlabel('Frame'),
plt.ylabel('Detection time (ms)'),
plt.title('Model ONNX chua toi uu')
plt.hlines(y=timeRecMean,
xmin=0,
xmax=frame_count,
linewidth=2,
color='r',
label='Mean')
plt.savefig('non-optimized.png', bbox_inches='tight')
plt.legend(loc='upper center',
bbox_to_anchor=(0.5, 0.5),
shadow=True,
ncol=2)
plt.show()
cap.release()
cv2.destroyAllWindows()
else:
raise TypeError('Unsupported File Format.')