def textRecognize(mask1,mask2,angleError,truAz,truEl):
# crop region of interest
roi1 = frame[mask1]
roi2 = frame[mask2]
# Convert to grayscale
roi1 = cv2.cv2Color(roi1, cv2.COLOR_BGR2GRAY)
roi2 = cv2.cv2Color(roi2, cv2.COLOR_BGR2GRAY)
# double image size, linear interpolation
roi1 = cv2.resize(roi1,None,fx=2,fy=2,interpolation=cv2.INTER_LINEAR)
roi2 = cv2.resize(roi2,None,fx=2,fy=2,interpolation=cv2.INTER_LINEAR)
# perform thresholding
level,roi1 = cv2.threshold(roi1,0,255,cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
level,roi2 = cv2.threshold(roi2,0,255,cv2.THRESH_BINARY_INV+cv2.THRESH_OTSU)
# Take ROI and use tesseract to convert to string
config = '--oem 1 --psm 7'
text1 = pytesseract.image_to_string(roi1,config=config)
text2 = pytesseract.image_to_string(roi2,config=config)
try:
textNum1 = int(text1[1:3]) + int(text1[4:6])/60 - truAz
textNum2 = int(text2[1:2]) + int(text1[3:5])/60 - truEl
textDiff = round(math.sqrt(textNum1**2 + textNum2**2),2)
pixel2truth = round(textDiff - angleError,2)
print(pixel2truth)
except:
print('failed')
def apply_portrait(frame):
frame = apply_alpha_convert(frame)
gray = cv2.cv2Color(frame, cv2.COLOR_BGR2GRAY)
_, mask = cv2.threshold(gray, 120, 255, cv2.THRESH_BINARY)
mask = cv2.cvtColor(mask, cv2.COLOR_GRAY2BGRA)
blurred = cv2.GaussianBlur(frame, (21, 21), 0)
blended = apply_blend(frame, blurred, mask)
frame = cv2.cvtColor(blended, cv2.COLOR_BGRA2BGR)
return frame
def load_image(filepath):
"""
Load image in dataset into numpy array
"""
image = cv2.imread(filepath)
# As cv2 loads images as BGR, convert to RGB
image = cv2.cv2Color(image, cv2.COLOR_BGR2RGB)
image = image.astype(np.float32)
return image
def strokeEdges(src, dst, blurKsize=7, edgeKsize=5):
if blurKsize >= 3: # 模糊核大小 > 3
blurredSrc = cv2.medianBlur(src, blurKsize) # 中值模糊
graySrc = cv2.cvtColor(blurredSrc, cv2.COLOR_BGR2GRAY) # 灰度化
else:
graySrc = cv2.cv2Color(src, cv2.COLOR_BGR2GRAY) # 直接灰度化
cv2.Laplacian(graySrc, cv2.CV_8U, graySrc, ksize=edgeKsize) # 边缘化
normalizedInverseAlpha = (1.0 / 255) * (255 - graySrc) # 用于反相并归一化
channels = cv2.split(src) # 分离通道
for channel in channels:
channel[:] = channel * normalizedInverseAlpha # 对每个通道的所有像素进行反相和归一化
cv2.merge(channels, dst) # 合并处理后的所有通道
def findArucoMarkers(img, markerSize=6, totalMarkers=250, draw=True):
imgGray = cv2.cv2Color(img, cv2.COLOR_BGR2GRAY)
key = getattr(aruco, f'DICT_{markerSize}*{markerSize}_{totalMarkers}')
arucoDict = aruco.Dictionary_get(key)
arucoParam = aruco.DetectorParameters_create()
bboxs, ids, rejected = aruco.detectMarkers(imgGray,
arucoDict,
parameters=arucoParam)
if draw:
aruco.drawDetectedMarkers(img, bboxs)
return [bboxs, ids]
def preprocess_frame(self, frame):
"""
Take the gradients of the frame
"""
image = cv2.imread(args["image"]) # read image from given path
gray = cv2.cv2Color(image, cv2.COLOR_BGR2GRAY)
ddepth = cv2.cv.CV_32F if imutils.is_cv2() else cv2.CV_32F
gradX = cv2.Sobel(gray, ddpeth=ddepth, dx=1, dy=0, ksize=-1)
gradY = cv2.Sobel(gray, ddepth=ddepth, dx=0, dy=1, ksize=-1)
gradient = cv2.subtract(gradX, gradY)
return gray
def LFDecodeLensletImageSimple(LensletImage, WhiteImage, LensletGridModel, DecodeOptions,
return_LFWeight=False, return_DecodeOptions=False, return_DebayerLensletImage=False,
return_CorrectedLensletImage=False):
# set default parameters
DecodeOptions["LevelLimits"] = [np.min(WhiteImage), np.max(WhiteImage)]
DecodeOptions["ResampMethod"] = "fast"
DecodeOptions["Precision"] = np.float32
DecodeOptions["DoDehex"] = True
DecodeOptions["DoSquareST"] = True
intmax_uint16 = float(2**16-1)
# Rescale image values, remove black level
DecodeOptions["LevelLimits"] = DecodeOptions["Precision"](
DecodeOptions["LevelLimits"])
BlackLevel = DecodeOptions["LevelLimits"][0]
WhiteLevel = DecodeOptions["LevelLimits"][1]
WhiteImage = WhiteImage.astype(DecodeOptions["Precision"])
WhiteImage = (WhiteImage - BlackLevel)/(WhiteLevel - BlackLevel)
LensletImage = LensletImage.astype(DecodeOptions["Precision"])
LensletImage = (LensletImage - BlackLevel)/(WhiteLevel - BlackLevel)
LensletImage = LensletImage/WhiteImage
# Clip -- this is aggressive and throws away bright areas; there is a potential for an HDR approach here
LensletImage = np.minimum(1, np.maximum(0, LensletImage))
nargout = 1 + np.sum([return_CorrectedLensletImage,
return_DebayerLensletImage, return_DecodeOptions, return_LFWeight])
if nargout < 2:
del WhiteImage
LensletImage = (
LensletImage*intmax_uint16).astype(DecodeOptions["Precision"])
# THIS LINE IS NOT SURE TO WORK. FIND SOMETHING CONCRETE THAT WORKS WITH LFS
LensletImage = cv2.cv2Color(LensletImage, cv2.COLOR_BAYER_BG2RGB)
LensletImage = LensletImage.astype(DecodeOptions["Precision"])
LensletImage = LensletImage/intmax_uint16
DecodeOptions["NColChans"] = 3
if nargout >= 2:
DecodeOptions["NWeightChans"] = 1
else:
DecodeOptions["NWeightChans"] = 0
if nargout > 3:
DebayerLensletImage = LensletImage
# Tranform to an integer-spaced grid
print("\nAligning image to lenslet array...")
InputSpacing = np.array(
[LensletGridModel["HSpacing"], LensletGridModel["VSpacing"]])
NewLensletSpacing = np.ceil(InputSpacing)
# Force even so hex shift is a whole pixel multiple
NewLensletSpacing = np.ceil(NewLensletSpacing/2)*2
XformScale = NewLensletSpacing/InputSpacing
NewOffset = np.array([LensletGridModel["HOffset"],
LensletGridModel["VOffset"]]) * XformScale
RoundedOffset = np.round(NewOffset)
XformTrans = RoundedOffset-NewOffset
NewLensletGridModel = {'HSpacing': NewLensletSpacing[0], 'VSpacing': NewLensletSpacing[1],
'HOffset': RoundedOffset[0], 'VOffset': RoundedOffset[1], 'Rot': 0,
'UMax': LensletGridModel["UMax"], 'VMax': LensletGridModel["VMax"],
'Orientation': LensletGridModel["Orientation"],
'FirstPosShiftRow': LensletGridModel["FirstPosShiftRow"]}
RRot = LFRotz(LensletGridModel["Rot"])
RScale = np.eye(3)
RScale[0, 0] = XformScale[0]
RScale[1, 1] = XformScale[1]
DecodeOptions.OutputScale[:2] = XformScale
DecodeOptions.OutputScale[2:4] = np.array([1, 2/np.sqrt(3)])
RTrans = np.eye(3)
RTrans[-1, :2] = XformTrans
# Change this later!!
''' This part of the code is not complete '''
FixAll = maketform('affine', RRot@RScale@RTrans)
temp = XformScale[0]
NewSize = np.shape(LensletImage[:, :, 0]*XformScale[::-1].reshape(1, 2))
LensletImage = imtransform(LensletImage, FixAll, 'YData', np.array(
[1, NewSize[0]]), 'XData', np.array([1, NewSize[1]]))
if nargout >= 2:
WhiteImage = imtransform(WhiteImage, FixAll, 'YData', np.array(
[1, NewSize[0]]), 'XData', np.array([1, NewSize[1]]))
if nargout >= 4:
CorrectedLensletImage = LensletImage
LF = SliceXYImage(NewLensletGridModel, LensletImage,
WhiteImage, DecodeOptions)
del WhiteImage, LensletImage
''' Till this much '''
# Correct for hex grid and resize to square u,v pixels
LFSize = list(np.shape(LF))
HexAspect = 2/np.sqrt(3)
if DecodeOptions["ResampMethod"] == "fast":
print("\nResampling (1D approximation) to square u,v pixels")
n_steps = int(np.ceil(LFSize[3]+1))
NewUVec = HexAspect*np.arange(n_steps)
NewUVec = NewUVec[:int(np.ceil(LFSize[3]*HexAspect))]
OrigUSize = LFSize[3]
LFSize[3] = len(NewUVec)
# Allocate dest and copy orig LF into it (memory saving vs. keeping both separately)
LF2 = np.zeros(LFSize, dtype=DecodeOptions["Precision"])
LF2[:, :, :, :OrigUSize, :] = LF
LF = LF2
del LF2
if DecodeOptions["DoDehex"]:
ShiftUVec = -0.5+NewUVec
print('removing hex sampling...')
else:
ShiftUVec = NewUVec
print("...")
for ColChan in range(np.shape(LF)[4]):
CurUVec = ShiftUVec
for RowIter in range(2):
# Check if this works!!!
RowIdx = np.mod(
NewLensletGridModel["FirstPosShiftRow"] + RowIter, 2) + 1
ShiftRows = np.squeeze(
LF[:, :, RowIdx:-1:2, :OrigUSize, ColChan])
SliceSize = list(np.shape(ShiftRows))
SliceSize[3] = len(NewUVec)
ShiftRows = ShiftRows.reshape(
SliceSize[0]*SliceSize[1]*SliceSize[2], np.shape(ShiftRows)[3])
ShiftRows_func = interp1d(
np.arange(np.shape(ShiftRows)[1]), ShiftRows, kind='linear')
ShiftRows = ShiftRows_func(CurUVec)
ShiftRows[~np.isfinite(ShiftRows)] = 0
LF[:, :, RowIdx:-1:2, :,
ColChan] = ShiftRows.reshape(SliceSize)
CurUVec = NewUVec
del ShiftRows, ShiftRows_func
DecodeOptions["OutputScale"][2] = DecodeOptions["OutputScale"][2] * HexAspect
elif DecodeOptions["ResampMethod"] == "triangulation":
pass
else:
print('\nNo valid dehex / resampling selected\n')
# Resize to square s,t pixels
# Assumes only a very slight resampling is required, resulting in an identically-sized output light field
if DecodeOptions["DoSquareST"]:
print('\nResizing to square s,t pixels using 1D linear interp...')
ResizeScale = DecodeOptions["OutputScale"][0] / \
DecodeOptions["OutputScale"][1]
ResizeDim1 = 0
ResizeDim2 = 1
if ResizeScale < 1:
ResizeScale = 1/ResizeScale
ResizeDim1 = 1
ResizeDim2 = 0
OrigSize = np.shape(LF)[ResizeDim1]
OrigVec = np.arange(OrigSize) - OrigSize//2
NewVec = OrigVec/ResizeScale
OrigDims = np.arange(5)
OrigDims = np.delete(OrigDims, ResizeDim1)
UBlkSize = 32
USize = np.shape(LF)[3]
# grab paths from args
imagePaths = sorted(glob.glob(args["images"] + "/*.png"))
maskPaths = sorted(glob.glob(args["masks"] + "/*.png"))
data = []
target = []
# yields 512 dimensional feature vector used to
# characterize the color of the flower
desc = RGBHistogram([8, 8, 8])
for (imagePath, maskPath) in zip(imagePaths, maskPaths):
image = cv2.imread(imagePath)
mask = cv2.imread(maskPath)
# convert to grayscale
mask = cv2.cv2Color(mask, cv2.COLOR_BGR2GRAY)
features = desc.describe(image, mask)
data.append(features)
target.append(imagePath.split("_")[-2])
# encode labels
# unique finds unique species names, which are fed to labelencoder
targetNames = np.unique(target)
le = LabelEncoder()
# fits unique species names into integers, a category for each species
# then transforms the strings into their corresponding integer classes
# target now contains list of integers, one for each data point
# where integer maps to flower species name
target = le.fit_transform(target)
def main():
#################
# configurations
#################
parser = argparse.ArgumentParser()
parser.add_argument("--input_path", type=str, required=True)
parser.add_argument("--gt_path", type=str, required=True)
parser.add_argument("--output_path", type=str, required=True)
parser.add_argument("--model_path", type=str, required=True)
parser.add_argument("--gpu_id", type=str, required=True)
#parser.add_argument("--screen_notation", type=str, required=True)
parser.add_argument('--opt', type=str, required=True, help='Path to option YAML file.')
args = parser.parse_args()
opt = option.parse(args.opt, is_train=False)
PAD = 32
total_run_time = AverageMeter()
print("GPU ", torch.cuda.device_count())
os.environ['CUDA_VISIBLE_DEVICES'] = str(args.gpu_id)
device = torch.device('cuda')
data_mode = 'sharp_bicubic'
flip_test = False
Input_folder = args.input_path
GT_folder = args.gt_path
Result_folder = args.output_path
Model_path = args.model_path
# create results folder
if not os.path.exists(Result_folder):
os.makedirs(Result_folder, exist_ok=True)
model_path = Model_path
N_in = 5
model = EDVR_arch.EDVR(nf=opt['network_G']['nf'], nframes=opt['network_G']['nframes'],
groups=opt['network_G']['groups'], front_RBs=opt['network_G']['front_RBs'],
back_RBs=opt['network_G']['back_RBs'],
predeblur=opt['network_G']['predeblur'], HR_in=opt['network_G']['HR_in'],
w_TSA=opt['network_G']['w_TSA'])
#### dataset
test_dataset_folder = Input_folder
GT_dataset_folder = GT_folder
#### evaluation
crop_border = 0
border_frame = N_in // 2 # border frames when evaluate
# temporal padding mode
padding = 'new_info'
save_imgs = True
save_folder = os.path.join(Result_folder, data_mode)
util.mkdirs(save_folder)
util.setup_logger('base', save_folder, 'test', level=logging.INFO, screen=True, tofile=True)
logger = logging.getLogger('base')
#### log info
logger.info('Data: {} - {}'.format(data_mode, test_dataset_folder))
logger.info('Padding mode: {}'.format(padding))
logger.info('Model path: {}'.format(model_path))
logger.info('Save images: {}'.format(save_imgs))
logger.info('Flip test: {}'.format(flip_test))
#### set up the models
model.load_state_dict(torch.load(model_path), strict=True)
model.eval()
model = model.to(device)
avg_psnr_l, avg_psnr_center_l, avg_psnr_border_l = [], [], []
avg_rgb_psnr_l, avg_rgb_psnr_center_l, avg_rgb_psnr_border_l = [], [], []
subfolder_name_l = []
subfolder_l = sorted(glob.glob(osp.join(test_dataset_folder, '*')))
subfolder_GT_l = sorted(glob.glob(osp.join(GT_dataset_folder, '*')))
end = time.time()
for subfolder in subfolder_l:
input_subfolder = os.path.split(subfolder)[1]
subfolder_GT = os.path.join(GT_dataset_folder, input_subfolder)
if not os.path.exists(subfolder_GT):
continue
print("Evaluate Folders: ", input_subfolder)
subfolder_name = osp.basename(subfolder)
subfolder_name_l.append(subfolder_name)
save_subfolder = osp.join(save_folder, subfolder_name)
img_path_l = sorted(glob.glob(osp.join(subfolder, '*')))
max_idx = len(img_path_l)
if save_imgs:
util.mkdirs(save_subfolder)
#### read LQ and GT images, notice we load yuv img here
imgs_LQ = data_util.read_img_seq_yuv(subfolder) # Num x 3 x H x W
img_GT_l = []
for img_GT_path in sorted(glob.glob(osp.join(subfolder_GT, '*'))):
img_GT_l.append(data_util.read_img_yuv(None, img_GT_path))
avg_psnr, avg_psnr_border, avg_psnr_center, N_border, N_center = 0, 0, 0, 0, 0
avg_rgb_psnr, avg_rgb_psnr_border, avg_rgb_psnr_center = 0, 0, 0
# process each image
for img_idx, img_path in enumerate(img_path_l):
img_name = osp.splitext(osp.basename(img_path))[0]
select_idx = data_util.index_generation(img_idx, max_idx, N_in, padding=padding)
imgs_in = imgs_LQ.index_select(0, torch.LongTensor(select_idx)).unsqueeze(0).to(device) # 960 x 540
# here we split the input images 960x540 into 9 320x180 patch
gtWidth = 3840
gtHeight = 2160
intWidth_ori = imgs_in.shape[4] # 960
intHeight_ori = imgs_in.shape[3] # 540
scale = 4
intPaddingRight = PAD # 32# 64# 128# 256
intPaddingLeft = PAD # 32#64 #128# 256
intPaddingTop = PAD # 32#64 #128#256
intPaddingBottom = PAD # 32#64 # 128# 256
pader = torch.nn.ReplicationPad2d([intPaddingLeft, intPaddingRight, intPaddingTop, intPaddingBottom])
imgs_in = torch.squeeze(imgs_in, 0) # N C H W
imgs_in = pader(imgs_in) # N C 604 1024
# todo: output 4k
X0 = imgs_in
X0 = torch.unsqueeze(X0, 0)
if flip_test:
output = util.flipx4_forward(model, X0)
else:
output = util.single_forward(model, X0)
# todo remove padding
output = output[0, :, intPaddingTop * scale:(intPaddingTop + intHeight_ori) * scale,
intPaddingLeft * scale: (intPaddingLeft + intWidth_ori) * scale]
output = util.tensor2img(output.squeeze(0))
print("*****************current image process time \t " + str(
time.time() - end) + "s ******************")
total_run_time.update(time.time() - end, 1)
# calculate PSNR on YUV
y_all = output / 255.
GT = np.copy(img_GT_l[img_idx])
y_all, GT = util.crop_border([y_all, GT], crop_border)
crt_psnr = util.calculate_psnr(y_all * 255, GT * 255)
logger.info('{:3d} - {:25} \tYUV_PSNR: {:.6f} dB'.format(img_idx + 1, img_name, crt_psnr))
# here we also calculate PSNR on RGB
y_all_rgb = data_util.ycbcr2rgb(output / 255.)
GT_rgb = data_util.ycbcr2rgb(np.copy(img_GT_l[img_idx]))
y_all_rgb, GT_rgb = util.crop_border([y_all_rgb, GT_rgb], crop_border)
crt_rgb_psnr = util.calculate_psnr(y_all_rgb * 255, GT_rgb * 255)
logger.info('{:3d} - {:25} \tRGB_PSNR: {:.6f} dB'.format(img_idx + 1, img_name, crt_rgb_psnr))
if save_imgs:
im_out = np.round(y_all_rgb*255.).astype(numpy.uint8)
# todo, notice here we got rgb img, but cv2 need bgr when saving a img
cv2.imwrite(osp.join(save_subfolder, '{}.png'.format(img_name)), cv2.cv2Color(im_out, cv2.COLOR_RGB2BGR))
# for YUV and RGB, respectively
if img_idx >= border_frame and img_idx < max_idx - border_frame: # center frames
avg_psnr_center += crt_psnr
avg_rgb_psnr_center += crt_rgb_psnr
N_center += 1
else: # border frames
avg_psnr_border += crt_psnr
avg_rgb_psnr_border += crt_rgb_psnr
N_border += 1
# for YUV
avg_psnr = (avg_psnr_center + avg_psnr_border) / (N_center + N_border)
avg_psnr_center = avg_psnr_center / N_center
avg_psnr_border = 0 if N_border == 0 else avg_psnr_border / N_border
avg_psnr_l.append(avg_psnr)
avg_psnr_center_l.append(avg_psnr_center)
avg_psnr_border_l.append(avg_psnr_border)
logger.info('Folder {} - Average YUV PSNR: {:.6f} dB for {} frames; '
'Center YUV PSNR: {:.6f} dB for {} frames; '
'Border YUV PSNR: {:.6f} dB for {} frames.'.format(subfolder_name, avg_psnr,
(N_center + N_border),
avg_psnr_center, N_center,
avg_psnr_border, N_border))
# for RGB
avg_rgb_psnr = (avg_rgb_psnr_center + avg_rgb_psnr_border) / (N_center + N_border)
avg_rgb_psnr_center = avg_rgb_psnr_center / N_center
avg_rgb_psnr_border = 0 if N_border == 0 else avg_rgb_psnr_border / N_border
avg_rgb_psnr_l.append(avg_rgb_psnr)
avg_rgb_psnr_center_l.append(avg_rgb_psnr_center)
avg_rgb_psnr_border_l.append(avg_rgb_psnr_border)
logger.info('Folder {} - Average RGB PSNR: {:.6f} dB for {} frames; '
'Center RGB PSNR: {:.6f} dB for {} frames; '
'Border RGB PSNR: {:.6f} dB for {} frames.'.format(subfolder_name, avg_rgb_psnr,
(N_center + N_border),
avg_rgb_psnr_center, N_center,
avg_rgb_psnr_border, N_border))
logger.info('################ Tidy Outputs ################')
# for YUV
for subfolder_name, psnr, psnr_center, psnr_border in zip(subfolder_name_l, avg_psnr_l, avg_psnr_center_l, avg_psnr_border_l):
logger.info('Folder {} - Average YUV PSNR: {:.6f} dB. '
'Center YUV PSNR: {:.6f} dB. '
'Border YUV PSNR: {:.6f} dB.'.format(subfolder_name, psnr, psnr_center, psnr_border))
# for RGB
for subfolder_name, psnr, psnr_center, psnr_border in zip(subfolder_name_l, avg_rgb_psnr_l, avg_rgb_psnr_center_l, avg_rgb_psnr_border_l):
logger.info('Folder {} - Average RGB PSNR: {:.6f} dB. '
'Center RGB PSNR: {:.6f} dB. '
'Border RGB PSNR: {:.6f} dB.'.format(subfolder_name, psnr, psnr_center, psnr_border))
logger.info('################ Final Results ################')
logger.info('Data: {} - {}'.format(data_mode, test_dataset_folder))
logger.info('Padding mode: {}'.format(padding))
logger.info('Model path: {}'.format(model_path))
logger.info('Save images: {}'.format(save_imgs))
logger.info('Flip test: {}'.format(flip_test))
logger.info('Total Average YUV PSNR: {:.6f} dB for {} clips. '
'Center YUV PSNR: {:.6f} dB. Border YUV PSNR: {:.6f} dB.'.format(
sum(avg_psnr_l) / len(avg_psnr_l), len(subfolder_l),
sum(avg_psnr_center_l) / len(avg_psnr_center_l),
sum(avg_psnr_border_l) / len(avg_psnr_border_l)))
logger.info('Total Average RGB PSNR: {:.6f} dB for {} clips. '
'Center RGB PSNR: {:.6f} dB. Border RGB PSNR: {:.6f} dB.'.format(
sum(avg_rgb_psnr_l) / len(avg_rgb_psnr_l), len(subfolder_l),
sum(avg_rgb_psnr_center_l) / len(avg_rgb_psnr_center_l),
sum(avg_rgb_psnr_border_l) / len(avg_rgb_psnr_border_l)))
import cv2
import numpy as np
cap = cv2.VideoCapture('vtest.avi')
ret, frame1 = cap.read()
ret, frame2 = cap.read()
while cap.isOpened():
diff = cv2.absdiff(frame1, frame2)
gray = cv2.cv2Color(diff, cv2.COLOR_BGR2GRAY)
blur = cv2.GaussianBlur(gray, (5, 5), 0)
_, thresh = cv2.threshold(blur, 20, 255)
ret, frame = cap.read()
cv2.imshow("inter", frame)
if cv2.waitKey(40) == 27:
break
cv2.destroyAllWindows()
cap.release()
