-
Notifications
You must be signed in to change notification settings - Fork 5
/
tpu_test.py
64 lines (46 loc) · 1.68 KB
/
tpu_test.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
import cv2
import argparse
import numpy as np
from modules import tpu
def parser():
p = argparse.ArgumentParser()
p.add_argument('-m', type=str)
p.add_argument('-v', type=str)
# quantization mean and standard deviation
p.add_argument('--quantmean', type=int, default=tpu.TPUParamDefaults.QMEAN)
p.add_argument('--quantstd', type=float, default=tpu.TPUParamDefaults.QSTD)
# dequantization mean and standard deviation
p.add_argument('--dequantmean', type=int, default=tpu.TPUParamDefaults.DQMEAN)
p.add_argument('--dequantstd', type=float, default=tpu.TPUParamDefaults.DQSTD)
return p.parse_args()
def main():
args = parser()
cap = cv2.VideoCapture(args.v)
runtime = tpu.TPUBenchTest(model=args.m)
while True:
ret, frame = cap.read()
if not ret:
# no frame retrieved
break
frame = np.array(frame)
frmcpy = frame.copy()
# quantize model input, run inference and dequantize outputs
frame = runtime.preprocess(frame, mean=args.quantmean, std=args.quantstd)
pred_obj = runtime.invoke(frame)
print('[Inference time] {}ms'.format(runtime.inference_time))
pred = runtime.postprocess(
pred_obj=pred_obj,
frame=frmcpy,
mean=args.dequantmean,
std=args.dequantstd
)
# resize output frame to fit foo
out = cv2.resize(pred, (pred.shape[1] // 2, pred.shape[0] // 2))
cv2.imshow('TPU Benchmark', out)
if cv2.waitKey(1) and 0xFF == ord('q'):
# exit on key press
break
cap.release()
cv2.destroyAllWindows()
if __name__ == "__main__":
main()