def run():
csv_file_path = args.csv_file_path
model_path = args.model_dir
precision = args.precision
# System Check
system_check = utilities(jetson_devkit=args.jetson_devkit,
gpu_freq=args.gpu_freq,
dla_freq=args.dla_freq)
if system_check.check_trt():
sys.exit()
system_check.set_power_mode(args.power_mode, args.jetson_devkit)
system_check.clear_ram_space()
system_check.run_set_clocks_withDVFS()
system_check.set_jetson_fan(255)
benchmark_data = read_write_data(csv_file_path=csv_file_path,
model_path=model_path)
# Actually run the ssd-mobilenet-v1 benchmark here
model = run_benchmark_models(csv_file_path=csv_file_path,
model_path=model_path,
precision=precision,
benchmark_data=benchmark_data)
download_err = model.execute(read_index=7)
if not download_err:
fps, error_log = model.report()
if not error_log:
model.remove()
system_check.clear_ram_space()
system_check.set_jetson_fan(0)
return fps[3]
def run():
csv_file_path = args.csv_file_path
model_path = args.model_dir
precision = args.precision
# System Check
system_check = utilities(jetson_devkit=args.jetson_devkit,
gpu_freq=args.gpu_freq,
dla_freq=args.dla_freq)
if system_check.check_trt():
sys.exit()
system_check.set_power_mode(args.power_mode, args.jetson_devkit)
system_check.clear_ram_space()
system_check.run_set_clocks_withDVFS()
system_check.set_jetson_fan(255)
benchmark_data = read_write_data(csv_file_path=csv_file_path,
model_path=model_path)
# Run through all the benchmarks here
running_total = 0
running_total += inception_benchmark.run()
running_total += vgg_benchmark.run()
running_total += superres_benchmark.run()
running_total += unet_benchmark.run()
running_total += pose_benchmark.run()
running_total += yolo_benchmark.run()
running_total += resnet_benchmark.run()
running_total += mobilenet_benchmark.run()
average_fps = running_total / 8
system_check.clear_ram_space()
system_check.set_jetson_fan(0)
return average_fps
def main():
camera = cv2.VideoCapture(0)
i = 0
util_funcs = utilities()
while True:
return_value, image = camera.read()
cv2.imwrite('temp/'+'opencv'+str(i)+'.png', image)
test_images,x1s,y1s,widths,heights = util_funcs.extract_multiple_faces('temp/'+'opencv'+str(i)+'.png')
test_image = 'temp/'+'opencv'+str(i)+'.png'
util_funcs.predict(test_image,test_images,x1s,y1s,widths,heights)
cv2.imwrite('temp/'+'opencv'+str(i)+'.png', image)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
i+=1
camera.release()
cv2.destroyAllWindows()
def main():
# System Check
arg_parser = setup_argparser()
args = arg_parser.make_args()
system_check = utilities(jetson_devkit=args.jetson_devkit,
gpu_freq=args.gpu_freq,
dla_freq=args.dla_freq)
if args.benchmark_mode == 1:
print("Set Jetson Benchmark Mode : ON!\n")
system_check.close_all_notice()
if system_check.check_trt():
sys.exit()
system_check.set_power_mode(args.power_mode, args.jetson_devkit)
system_check.clear_ram_space()
if args.jetson_clocks:
system_check.set_jetson_clocks()
else:
system_check.run_set_clocks_withDVFS()
system_check.set_jetson_fan(255)
else:
system_check.clear_ram_space()
system_check.set_jetson_fan(0)
print("Set Jetson Benchmark Mode : OFF!\n")
def main():
# Set Parameters
arg_parser = benchmark_argparser()
args = arg_parser.make_args()
csv_file_path = args.csv_file_path
model_path = args.model_dir
precision = args.precision
dummy_check = utilities()
if dummy_check.check_trt():
print(
"Couldn't find TensorRT, please check trtexec in the path(\"/usr/src/tensorrt/bin/trtexec \")"
)
sys.exit()
dummy_check.close_all_warning()
# Read CSV and Write Data
benchmark_data = read_write_data(csv_file_path=csv_file_path,
model_path=model_path)
if args.all:
latency_each_model = []
print(
"Running all benchmarks.. This would take long time if the trt engine is not prepared..."
)
for read_index in range(0, len(benchmark_data)):
gc.collect()
model = run_benchmark_models(csv_file_path=csv_file_path,
model_path=model_path,
precision=precision,
benchmark_data=benchmark_data)
download_err = model.execute(read_index=read_index)
if not download_err:
# Reading Results
latency_fps, error_log = model.report()
latency_each_model.append(latency_fps)
del gc.garbage[:]
benchmark_table = pd.DataFrame(latency_each_model,
columns=[
'GPU (ms)', 'DLA0 (ms)',
'DLA1 (ms)', 'FPS', 'Model Name'
],
dtype=float)
# Note: GPU, DLA latencies are measured in miliseconds, FPS = Frames per Second
print(benchmark_table[['Model Name', 'FPS']])
if args.plot:
benchmark_data.plot_perf(latency_each_model)
elif args.model_name == 'inception_v4':
model = run_benchmark_models(csv_file_path=csv_file_path,
model_path=model_path,
precision=precision,
benchmark_data=benchmark_data)
download_err = model.execute(read_index=0)
if not download_err:
_, error_log = model.report()
elif args.model_name == 'super_resolution':
model = run_benchmark_models(csv_file_path=csv_file_path,
model_path=model_path,
precision=precision,
benchmark_data=benchmark_data)
download_err = model.execute(read_index=2)
if not download_err:
_, error_log = model.report()
elif args.model_name == 'unet':
model = run_benchmark_models(csv_file_path=csv_file_path,
model_path=model_path,
precision=precision,
benchmark_data=benchmark_data)
download_err = model.execute(read_index=3)
if not download_err:
_, error_log = model.report()
elif args.model_name == 'tiny-yolov3':
model = run_benchmark_models(csv_file_path=csv_file_path,
model_path=model_path,
precision=precision,
benchmark_data=benchmark_data)
download_err = model.execute(read_index=5)
if not download_err:
_, error_log = model.report()
elif args.model_name == 'resnet':
model = run_benchmark_models(csv_file_path=csv_file_path,
model_path=model_path,
precision=precision,
benchmark_data=benchmark_data)
download_err = model.execute(read_index=6)
if not download_err:
_, error_log = model.report()
elif args.model_name == 'ssd-mobilenet-v1':
model = run_benchmark_models(csv_file_path=csv_file_path,
model_path=model_path,
precision=precision,
benchmark_data=benchmark_data)
download_err = model.execute(read_index=7)
if not download_err:
_, error_log = model.report()
def main():
# Set Parameters
arg_parser = benchmark_argparser()
args = arg_parser.make_args()
csv_file_path = args.csv_file_path
model_path = args.model_dir
precision = args.precision
# System Check
system_check = utilities(jetson_devkit=args.jetson_devkit, gpu_freq=args.gpu_freq, dla_freq=args.dla_freq)
system_check.close_all_apps()
if system_check.check_trt():
sys.exit()
system_check.set_power_mode(args.power_mode, args.jetson_devkit)
system_check.clear_ram_space()
if args.jetson_clocks:
system_check.set_jetson_clocks()
else:
system_check.run_set_clocks_withDVFS()
system_check.set_jetson_fan(255)
# Read CSV and Write Data
benchmark_data = read_write_data(csv_file_path=csv_file_path, model_path=model_path)
if args.all:
latency_each_model =[]
print("Running all benchmarks.. This will take at least 2 hours...")
for read_index in range (0,len(benchmark_data)):
gc.collect()
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=read_index)
if not download_err:
# Reading Results
latency_fps, error_log = model.report()
latency_each_model.append(latency_fps)
# Remove engine and txt files
if not error_log:
model.remove()
del gc.garbage[:]
system_check.clear_ram_space()
benchmark_table = pd.DataFrame(latency_each_model, columns=['GPU (ms)', 'DLA0 (ms)', 'DLA1 (ms)', 'FPS', 'Model Name'], dtype=float)
# Note: GPU, DLA latencies are measured in miliseconds, FPS = Frames per Second
print(benchmark_table[['Model Name', 'FPS']])
if args.plot:
benchmark_data.plot_perf(latency_each_model)
elif args.model_name == 'inception_v4':
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=0)
if not download_err:
_, error_log = model.report()
if not error_log:
model.remove()
elif args.model_name == 'vgg19':
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=1)
if not download_err:
_, error_log = model.report()
if not error_log:
model.remove()
elif args.model_name == 'super_resolution':
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=2)
if not download_err:
_, error_log = model.report()
if not error_log:
model.remove()
elif args.model_name == 'unet':
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=3)
if not download_err:
_, error_log = model.report()
if not error_log:
model.remove()
elif args.model_name == 'pose_estimation':
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=4)
if not download_err:
_, error_log = model.report()
if not error_log:
model.remove()
elif args.model_name == 'tiny-yolov3':
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=5)
if not download_err:
_, error_log = model.report()
if not error_log:
model.remove()
elif args.model_name == 'resnet':
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=6)
if not download_err:
_, error_log = model.report()
if not error_log:
model.remove()
elif args.model_name == 'ssd-mobilenet-v1':
model = run_benchmark_models(csv_file_path=csv_file_path, model_path=model_path, precision=precision, benchmark_data=benchmark_data)
download_err = model.execute(read_index=7)
if not download_err:
_, error_log = model.report()
if not error_log:
model.remove()
system_check.clear_ram_space()
system_check.set_jetson_fan(0)
