下面的代码将遍历HDD上具有 620,000帧的文件,我正在使用OpenCV的DNN面部检测器提取面部。它可以正常工作,但是每帧大约需要1秒= 172小时。 因此,我想使用多线程来加快速度,但是不确定如何做到这一点。
注意:我的笔记本电脑上有4个CPU内核,硬盘的读写速度约为100 MB / s
文件路径示例:/Volumes/HDD/frames/Fold1_part1/01/0/04541.jpg
frames_path = "/Volumes/HDD/frames"
path_HDD = "/Volumes/HDD/Data"
def filePath(path):
for root, directories, files in os.walk(path, topdown=False):
for file in files:
if (directories == []):
pass
elif (len(directories) > 3):
pass
elif (len(root) == 29):
pass
else:
# Only want the roots with /Volumes/HDD/Data/Fold1_part1/01
for dir in directories:
path_video = os.path.join(root, dir)
for r, d, f in os.walk(path_video, topdown=False):
for fe in f:
fullPath = r[:32]
label = r[-1:]
folds = path_video.replace("/Volumes/HDD/Data/", "")
finalPath = os.path.join(frames_path, folds)
finalImage = os.path.join(finalPath, fe)
fullImagePath = os.path.join(path_video, fe)
try :
if (os.path.exists(finalPath) == False):
os.makedirs(finalPath)
extractFaces(fullImagePath, finalImage)
except OSError as error:
print(error)
sys.exit(0)
def extractFaces(imageTest, savePath):
model = "/Users/yudhiesh/Downloads/deep-learning-face-detection/res10_300x300_ssd_iter_140000.caffemodel"
prototxt = "/Users/yudhiesh/Downloads/deep-learning-face-detection/deploy.prototxt.txt"
net = cv2.dnn.readNet(model, prototxt)
# load the input image and construct an input blob for the image
# by resizing to a fixed 300x300 pixels and then normalizing it
image = cv2.imread(imageTest)
(h, w) = image.shape[:2]
blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 1.0,(300, 300), (104.0, 177.0, 123.0))
print(f'Current file path {imageTest}')
# pass the blobs through the network and obtain the predictions
print("Computing object detections....")
net.setInput(blob)
detections = net.forward()
# Detect face with highest confidence
for i in range(0, detections.shape[2]):
box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
(startX, startY, endX, endY) = box.astype("int")
confidence = detections[0, 0, i, 2]
# If confidence > 0.5, save it as a separate file
if (confidence > 0.5):
frame = image[startY:endY, startX:endX]
rect = dlib.rectangle(startX, startY, endX, endY)
image = image[startY:endY, startX:endX]
print(f'Saving image to {savePath}')
cv2.imwrite(savePath, image)
if __name__ == "__main__":
filePath(path_HDD)
答案 0 :(得分:0)
设法将时间缩短为每张图像0.09-0.1秒。感谢您提出使用ProcessPoolExecutor的建议。
frames_path = "/Volumes/HDD/frames"
path_HDD = "/Volumes/HDD/Data"
def filePath(path):
for root, directories, files in os.walk(path, topdown=False):
for file in files:
if (directories == []):
pass
elif (len(directories) > 3):
pass
elif (len(root) == 29):
pass
else:
# Only want the roots with /Volumes/HDD/Data/Fold1_part1/01
for dir in directories:
path_video = os.path.join(root, dir)
for r, d, f in os.walk(path_video, topdown=False):
for fe in f:
fullPath = r[:32]
label = r[-1:]
folds = path_video.replace("/Volumes/HDD/Data/", "")
finalPath = os.path.join(frames_path, folds)
finalImage = os.path.join(finalPath, fe)
fullImagePath = os.path.join(path_video, fe)
try :
if (os.path.exists(finalPath) == False):
os.makedirs(finalPath)
with concurrent.futures.ProcessPoolExecutor() as executor:
executor.map(extractFaces(fullImagePath, finalImage))
except OSError as error:
print(error)
sys.exit(0)
def extractFaces(imageTest, savePath):
model = "/Users/yudhiesh/Downloads/deep-learning-face-detection/res10_300x300_ssd_iter_140000.caffemodel"
prototxt = "/Users/yudhiesh/Downloads/deep-learning-face-detection/deploy.prototxt.txt"
net = cv2.dnn.readNet(model, prototxt)
# load the input image and construct an input blob for the image
# by resizing to a fixed 300x300 pixels and then normalizing it
image = cv2.imread(imageTest)
(h, w) = image.shape[:2]
blob = cv2.dnn.blobFromImage(cv2.resize(image, (300, 300)), 1.0,(300, 300), (104.0, 177.0, 123.0))
print(f'Current file path {imageTest}')
# pass the blobs through the network and obtain the predictions
print("Computing object detections....")
net.setInput(blob)
detections = net.forward()
# Detect face with highest confidence
for i in range(0, detections.shape[2]):
box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
(startX, startY, endX, endY) = box.astype("int")
confidence = detections[0, 0, i, 2]
# If confidence > 0.5, save it as a separate file
if (confidence > 0.5):
frame = image[startY:endY, startX:endX]
rect = dlib.rectangle(startX, startY, endX, endY)
image = image[startY:endY, startX:endX]
print(f'Saving image to {savePath}')
cv2.imwrite(savePath, image)
if __name__ == "__main__":
filePath(path_HDD)