if args.resume:
print('==> Resuming from checkpoint..')
checkpoint = torch.load('./checkpoint/' + args.network)
net = checkpoint['net']
top1_acc = checkpoint['top1_acc']
top5_acc = checkpoint['top5_acc']
else:
net = mobilenetv2()
best_acc = 0
#For pruning and quantization
elif args.mode == 2:
checkpoint = torch.load('./checkpoint/' + args.network)
net = checkpoint['net']
params = utils.paramsGet(net)
thres = utils.findThreshold(params, args.pr)
mask_prune = utils.getPruningMask(net, thres)
if args.resume:
print('==> Resuming from checkpoint..')
best_acc = 0
else:
best_acc = 0
if use_cuda:
net.cuda()
net = torch.nn.DataParallel(net,
device_ids=range(torch.cuda.device_count()))
cudnn.benchmark = True
criterion = nn.CrossEntropyLoss()
blinks = []
nblinks = 0
blink_counter = 0
window_sum = 0
window_count = 0
# Writing the result in a video stream
out = cv2.VideoWriter('result.avi', cv2.VideoWriter_fourcc(*'XVID'), 29.0, (int(cap.get(3)), int(cap.get(4))))
df = pd.DataFrame()
print('thresholdng...')
while True:
s, frame = cap.read()
if frame is None: break
gray_f = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
gray_f = cv2.GaussianBlur(gray_f, (5, 5), 0) # to reduce noise
threshold, threshold_flag = findThreshold(gray_f)
if threshold_flag:
print(f'Threshold is done! \n threshold is {threshold}')
break
# Once thresholding is done, start processing the video on the calculated threshold
while True:
s, frame = cap.read()
if frame is None:
break
gray_f = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
gray_f = cv2.GaussianBlur(gray_f, (5, 5), 0)
h, w, = gray_f.shape