def train(self):
print("start training:")
# 注：必须要将unicode转成普通字符串，否则dlib无法读取，这个编码上的bug也是坑了我两个小时
train_xml_path = str(self.train_xml_path)
model_path = str(self.model_path)
# print(train_xml_path)
# print(model_path)
# print(chardet.detect(train_xml_path))
# xml = 'F:\\Python\\Programs\\my_dlib_face_detection_application\\images\\test_object_detector\\cats_train\\cat.xml'
# model = 'F:\\Python\\Programs\\my_dlib_face_detection_application\\images\\test_object_detector\\cats_train\\detector.svm'
# print(xml)
# print(model)
# print(chardet.detect(xml))
dlib.train_simple_object_detector(train_xml_path, model_path, self.options)
print("Training accuracy: {}".format(dlib.test_simple_object_detector(train_xml_path, model_path)))
print("The SVM model is saved in {0}".format(model_path))
if self.log:
self.log.append(u'--'*30)
self.log.append("Training complete!")
self.log.append("Training accuracy: {}".format(dlib.test_simple_object_detector(train_xml_path, model_path)))
self.log.append("The SVM model is saved in {0}".format(model_path))
self.log.append(u'--'*30)
def test_detector_accuracy(det_name):
folder = "../data/TrainHOG/"
coins_folder = folder
training_xml_path = os.path.join(coins_folder, "traincoins.xml")
testing_xml_path = os.path.join(coins_folder, "testcoins.xml")
detective = str(det_name)
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, detective)))
print("Testing accuracy: {}".format(
dlib.test_simple_object_detector(testing_xml_path, detective)))
def train_with_dlib(self, obj_folder, detector_path, train_file_name, test_file_name):
pdb.set_trace()
logger = MyUtils.Tee("{0}/{1}.log".format(obj_folder, 'run'), 'w')
logger.write('tee: dlib training called')
logger.write('dlib training called')
# Now let's do the training. The train_simple_object_detector() function has a
# bunch of options, all of which come with reasonable default values. The next
# few lines goes over some of these options.
options = dlib.simple_object_detector_training_options()
# The trainer is a kind of support vector machine and therefore has the usual
# SVM C parameter. In general, a bigger C encourages it to fit the training
# data better but might lead to overfitting. You must find the best C value
# empirically by checking how well the trained detector works on a test set of
# images you haven't trained on. Don't just leave the value set at 5. Try a
# few different C values and see what works best for your data.
options.C = 5
# Tell the code how many CPU cores your computer has for the fastest training.
options.num_threads = 2
options.be_verbose = True
training_xml_path = os.path.join(obj_folder, train_file_name)
testing_xml_path = os.path.join(obj_folder, test_file_name)
# This function does the actual training. It will save the final detector to
# detector.svm. The input is an XML file that lists the images in the training
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
logger.write('start training. saved detector path: ' + detector_path)
dlib.train_simple_object_detector(training_xml_path, detector_path, options)
logger.write( 'end training')
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will logger.write((the precision, recall, and then)
# average precision.
logger.write("") # Print blank line to create gap from previous output
logger.write("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, detector_path)))
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
accuracy = dlib.test_simple_object_detector(testing_xml_path, "detector.svm")
logger.write("Testing accuracy: {}".format(accuracy))
logger.flush()
return accuracy
def evaluate_dlib_detector(self):
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("MTB detector evaluation:")
print("Training accuracy: {}".format(dlib.test_simple_object_detector("../data/annotations/bb/training/combined_mtb.xml", "../data/models/detector_mtb.svm")))
print("Testing accuracy: {}".format(dlib.test_simple_object_detector("../data/annotations/bb/testing/combined_mtb.xml", "../data/models/detector_mtb.svm")))
print("PED detector evaluation:")
print("Training accuracy: {}".format(dlib.test_simple_object_detector("../data/annotations/bb/training/combined_ped.xml", "../data/models/detector_ped.svm")))
print("Testing accuracy: {}".format(dlib.test_simple_object_detector("../data/annotations/bb/testing/combined_ped.xml", "../data/models/detector_ped.svm")))
def evaluate_dlib_detector(self):
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("MTB detector evaluation:")
print("Training accuracy: {}".format(dlib.test_simple_object_detector("../data/training/training_mtb.xml", "../data/models/detector_mtb.svm")))
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
print("Testing accuracy: {}".format(dlib.test_simple_object_detector("../data/testing/testing_mtb.xml", "../data/models/detector_mtb.svm")))
print("PED detector evaluation:")
print("Training accuracy: {}".format(dlib.test_simple_object_detector("../data/training/training_ped.xml", "../data/models/detector_ped.svm")))
print("Testing accuracy: {}".format(dlib.test_simple_object_detector("../data/testing/testing_ped.xml", "../data/models/detector_ped.svm")))
def create_svm(request,pk,template_name='create_svm.html'):
brand = get_object_or_404(Brands, pk=pk)
data = {}
data[ 'pk' ] = pk
if request.method == 'POST':
svm = request.POST.get('svm')
cval = request.POST.get('cval')
paths = Paths.objects.filter( brand_id = pk )
if paths.exists():
for pt in paths:
train_path = pt.train_path
train_path = os.path.relpath(train_path,settings.MEDIA_ROOT)
test_path = pt.test_path
test_path = os.path.relpath(test_path,settings.MEDIA_ROOT)
svm_path = pt.svm_path
#cmd = "python 35Hawkdetect.py /var/sites/thirdauth/static/images/companies/Stovekraft\ Private\ Limited\(mohsin\)/idea/test/ cola123.svm jpg"
print svm_path
print settings.MEDIA_ROOT+"/"+train_path+"/"
faces_folder = settings.MEDIA_ROOT+"/"+train_path+"/"
test_folder = settings.MEDIA_ROOT+"/"+test_path+"/"
print faces_folder
#dest = "/var/sites/thirdauth/static/images/companies/Stovekraft Private Limited(mohsin)/Docomo/svm"
timestr = svm+time.strftime("_%m_%d_%H_%M")+".svm"
print timestr
outputpath = str(svm)+".svm"
cval = cval
options = dlib.simple_object_detector_training_options()
options.add_left_right_image_flips = True
options.C = int(cval)
options.num_threads = 4
options.be_verbose = True
training_xml_path = os.path.join(str(faces_folder), "training.xml")
testing_xml_path = os.path.join(str(test_folder), "training.xml")
dlib.train_simple_object_detector(training_xml_path,outputpath, options)
print("")
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, outputpath)))
print("Testing accuracy: {}".format(
dlib.test_simple_object_detector(testing_xml_path, outputpath)))
result = "Training accuracy: {}"+format(dlib.test_simple_object_detector(training_xml_path, outputpath))+" Testing accuracy: {}"+format(dlib.test_simple_object_detector(testing_xml_path, outputpath))
os.rename(str(outputpath),timestr)
if os.path.exists(str(svm_path)+"/"+str(timestr)):
os.remove(str(svm_path)+"/"+str(timestr))
shutil.move("/var/sites/thirdauth/"+str(timestr),str(svm_path))
Svms(svm_name = str(timestr), brand_id = pk , company_id = brand.company_id ).save()
return HttpResponse(result)
else:
return render(request,template_name, data ,context_instance=RequestContext(request))
def test_params(C, nuclear_norm):
options = dlib.simple_object_detector_training_options()
# our dataset is oriented, so definitely don't add in flips.
options.add_left_right_image_flips = False
options.C = C
options.num_threads = 1
options.be_verbose = False
options.nuclear_norm_regularization_strength = nuclear_norm
options.max_runtime_seconds = 5 # SET REALLY SMALL SO THE DEMO DOESN'T TAKE TO LONG, USE BIGGER VALUES FOR REAL USE!!!!!
dlib.train_simple_object_detector(
"images/small_face_dataset/cluster_001_train.xml", "detector1_.svm",
options)
# You can do a lot here. Run the detector through
# dlib.threshold_filter_singular_values() for instance to make sure it
# learns something that will work once thresholded. We can also add a
# penalty for having a lot of filters. Run this program a few times and
# try out different ways of penalizing the return from test_params() and
# see what happens.
result = dlib.test_simple_object_detector(
"images/small_face_dataset/cluster_001_test.xml", "detector1_.svm")
print("C = {}, nuclear_norm = {}".format(C, nuclear_norm))
print("testing accuracy: ", result)
sys.stdout.flush()
# For settings with the same average precision, we should prefer smaller C
# since smaller C has better generalization.
return result.average_precision - C * 1e-8
def do_train(self, training_xml_path, testing_xml_path, name_svm,
hyperparameters):
options = dlib.simple_object_detector_training_options()
options.detection_window_size = int(hyperparameters[0])
# options.add_left_right_image_flips = True
options.C = int(hyperparameters[2])
options.num_threads = int(hyperparameters[1])
options.epsilon = hyperparameters[3]
options.be_verbose = True
dlib.train_simple_object_detector(training_xml_path, name_svm, options)
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, name_svm)))
print("Testing accuracy: {}".format(
dlib.test_simple_object_detector(testing_xml_path, name_svm)))
self.detector = name_svm
def training_data(train_xml, test_xml, detectorName, Csvm=5):
options = dlib.simple_object_detector_training_options()
options.add_left_right_image_flips = True
options.C = Csvm
options.num_threads = 4
options.be_verbose = True
dlib.train_simple_object_detector(train_xml, detectorName + ".svm",
options)
print("Testing accuracy: {}".format(
dlib.test_simple_object_detector(test_xml, detectorName + ".svm")))
def train():
options = dlib.simple_object_detector_training_options()
options.add_left_right_image_flips = True
options.C = 5
options.num_threads = 4
options.be_verbose = True
options.detection_window_size = 1024
options.match_eps = 0.1
training_xml_path = "../pics/train/training-single.xml"
# training_xml_path = "/home/external/moderation-p**n-detector/oboobs.dlibxml"
dlib.train_simple_object_detector(training_xml_path, "../boobs.svm", options)
print("Training accuracy: {}".format(dlib.test_simple_object_detector(training_xml_path, "../boobs.svm")))
def train(training_xml_path, model_file="detector.svm"):
assert os.path.isfile(training_xml_path)
assert not os.path.isfile(model_file)
# Now let's do the training. The train_simple_object_detector() function has a
# bunch of options, all of which come with reasonable default values. The next
# few lines goes over some of these options.
options = dlib.simple_object_detector_training_options()
# Since faces are left/right symmetric we can tell the trainer to train a
# symmetric detector. This helps it get the most value out of the training
# data.
options.add_left_right_image_flips = True
# The trainer is a kind of support vector machine and therefore has the usual
# SVM C parameter. In general, a bigger C encourages it to fit the training
# data better but might lead to overfitting. You must find the best C value
# empirically by checking how well the trained detector works on a test set of
# images you haven't trained on. Don't just leave the value set at 5. Try a
# few different C values and see what works best for your data.
options.C = 10
# Tell the code how many CPU cores your computer has for the fastest training.
options.num_threads = 6
options.epsilon = 0.001
options.be_verbose = True
options.detection_window_size = 4096 #(32, 32)
# options.upsample_limit = 8
# This function does the actual training. It will save the final detector to
# detector.svm. The input is an XML file that lists the images in the training
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
print("Goingt to train ...")
dlib.train_simple_object_detector(training_xml_path, model_file, options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, model_file)))
def train():
options = dlib.simple_object_detector_training_options()
options.add_left_right_image_flips = True
options.C = 5
options.num_threads = 4
options.be_verbose = True
options.detection_window_size = 1024
options.match_eps = 0.1
training_xml_path = "../pics/train/training-single.xml"
# training_xml_path = "/home/external/moderation-p**n-detector/oboobs.dlibxml"
dlib.train_simple_object_detector(training_xml_path, "../boobs.svm",
options)
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, "../boobs.svm")))
def CREATE_SVM_DETECTOR():
options = dlib.simple_object_detector_training_options()
options.C = 5
options.num_threads = 4
options.be_verbose = True
# options.add_left_right_image_flips = True
fname_xml_train = 'data/xml/cat_BHS.xml'
# fname_xml_test = 'data/xml/testing.xml'
dlib.train_simple_object_detector(fname_xml_train, "cat_detector.svm",
options)
print()
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(fname_xml_train, "cat_detector.svm")))
def train(path_xml):
# objeto contenedor de las opciones para la rutina train_simple_object_detector()
# todas las opciones vienen con valores por defecto razonables
# http://dlib.net/python/index.html#dlib.simple_object_detector_training_options
options = dlib.simple_object_detector_training_options()
options.C = 6 # parametro C de las SVM, valores grandes pueden llevar al overfitting
options.add_left_right_image_flips = True # para objetos simetricos como las caras
options.be_verbose = True
options.epsilon = 0.005 # epsilon de detencion, valores pequeños -> accurate training
# utilizando la herramienta https://imglab.ml
dlib.train_simple_object_detector(path_xml, "detector.svm", options)
print("\nTraining accuracy: ",
dlib.test_simple_object_detector(path_xml, "detector.svm"))
def step4(self, btn):
#Based on dlib example:
# Now let's do the training. The train_simple_object_detector() function has a
# bunch of options, all of which come with reasonable default values. The next
# few lines goes over some of these options.
options = dlib.simple_object_detector_training_options()
# Since faces are left/right symmetric we can tell the trainer to train a
# symmetric detector. This helps it get the most value out of the training
# data.
options.add_left_right_image_flips = True
# The trainer is a kind of support vector machine and therefore has the usual
# SVM C parameter. In general, a bigger C encourages it to fit the training
# data better but might lead to overfitting. You must find the best C value
# empirically by checking how well the trained detector works on a test set of
# images you haven't trained on. Don't just leave the value set at 5. Try a
# few different C values and see what works best for your data.
options.C = 5
# Tell the code how many CPU cores your computer has for the fastest training.
options.num_threads = 4
options.be_verbose = True
trainingXML = os.path.join(self.tmp, 'training.xml')
#Ideally there would be half training, half testing:
testingXML = os.path.join(self.tmp, "training.xml")
# This function does the actual training. It will save the final detector to
# detector.svm. The input is an XML file that lists the images in the training
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes.
dlib.train_simple_object_detector(
trainingXML, os.path.join(self.tmp, "detector.svm"), options)
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(
trainingXML, os.path.join(self.tmp, "detector.svm"))))
def simple_training(trainpath="Training/", nbthreads=4, cvalue=5):
train_folder = trainpath
#Para treinarmos nosso dataset, chamaremos a classe train_simple_object_detector() com todos os seus valores default (que são razoavelmente bons)
options = dlib.simple_object_detector_training_options()
#Cartas são de certa forma simétricas, então, melhoraremos a qualidade do resultado mudando o padrão de add_left_right_image_flips para True
options.add_left_right_image_flips = True
#O valor de C encoraja um fitting melhor nos dados, mas um C muito grande encoraja overfitting, valor 5 ainda é experimental, precisamos de mais teste
options.C = cvalue
# Quantos Threads temos disponíveis para treinar em paralelo?
options.num_threads = nbthreads
#Vamos acompanhar o processo
options.be_verbose = True
#Concatene o diretório do treino e seu .xml correspondente numa string
training_xml_path = os.path.join(train_folder, "training.xml")
#testing_xml_path = os.path.join(train_folder, "testing.xml")
#Finalmente chamamos a função que faz o treino de fato. Salva o resultado em um detector .svm
# a partir do nosso .xml (Dados do treino supervisionado)
dlib.train_simple_object_detector(training_xml_path, "detector.svm",
options)
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, "detector.svm")))
print("Process done sucesssfully")
def fit(self, imagePaths, annotations, trainAnnot, trainImages, visualize=False, savePath=None):
st = time.time()
print(st)
annotations = self._prepare_annotations(annotations)
images = self._prepare_images(imagePaths)
###############################################
############### Entrenamiento
self._detector = dlib.train_simple_object_detector(images, annotations, self.options)
print('Entrenamiento completado. Tiempo: {:.2f} segundos'.format(time.time() - st))
#visualize HOG
if visualize:
win = dlib.image_window()
win.set_image(self._detector)
dlib.hit_enter_to_continue()
#Se guarda el detector y se evalua su desempeño usando las imagenes de prueba
if savePath is not None:
self._detector.save(savePath)
trainAnnot = self._prepare_annotations(trainAnnot)
trainImages = self._prepare_images(trainImages)
print("Metricas de entrenamiento : {}".format(dlib.test_simple_object_detector(trainImages,trainAnnot,self._detector)) )
return self
def training_cat():
import os
import sys
import glob
import dlib
import cv2
# options用于设置训练的参数和模式
options = dlib.simple_object_detector_training_options()
# Since faces are left/right symmetric we can tell the trainer to train a
# symmetric detector. This helps it get the most value out of the training
# data.
options.add_left_right_image_flips = True
# 支持向量机的C参数，通常默认取为5.自己适当更改参数以达到最好的效果
options.C = 5
# 线程数，你电脑有4核的话就填4
options.num_threads = 4
options.be_verbose = True
# 获取路径
current_path = os.getcwd()
train_xml_path = r'E:\bigdata\ai\cat\cat.xml'
print("training file path:" + train_xml_path)
# print(train_xml_path)
# 开始训练
print("start training:")
dlib.train_simple_object_detector(train_xml_path,
r'E:\bigdata\ai\cat\detector.svm',
options)
print("---------")
# 测试训练模型
print("Training accuracy（准确性）: {}".format(
dlib.test_simple_object_detector(train_xml_path,
r'E:\bigdata\ai\cat\detector.svm')))
def train_detector(train_data, filename='detector.svm'):
'''Trains an object detector (HOG + SVM) and saves the model'''
# Seperate the images and bounding boxes in different lists.
images = [val[0] for val in train_data.values()]
bounding_boxes = [val[1] for val in train_data.values()]
# Initialize object detector Options
options = dlib.simple_object_detector_training_options()
options.add_left_right_image_flips = False
options.C = 5
# Train the model
detector = dlib.train_simple_object_detector(images, bounding_boxes,
options)
# Check results
results = dlib.test_simple_object_detector(images, bounding_boxes,
detector)
print(f'Training Results: {results}')
# Save model
detector.save(filename)
print(f'Saved the model to {filename}')
import dlib
# construct the argument parse and parse the arguments
ap = argparse.ArgumentParser()
ap.add_argument("-x", "--xml", required=True, help="path to input XML file")
ap.add_argument("-d",
"--detector",
required=True,
help="Path to the object detector")
args = vars(ap.parse_args())
# grab the default training options for the HOG + Linear SVM detector, then
# train the detector -- in practice, the `C` parameter should be cross-validated
print("[INFO] training detector...")
options = dlib.simple_object_detector_training_options()
options.C = 1.0
options.num_threads = 4
options.be_verbose = True
options.upsample_limit = 1
dlib.train_simple_object_detector(args["xml"], args["detector"], options)
# show the training accuracy
print("[INFO] training accuracy: {}".format(
dlib.test_simple_object_detector(args["xml"], args["detector"])))
# load the detector and visualize the HOG filter
detector = dlib.simple_object_detector(args["detector"])
win = dlib.image_window()
win.set_image(detector)
dlib.hit_enter_to_continue()
# data.
options.add_left_right_image_flips = True
# The trainer is a kind of support vector machine and therefore has the usual
# SVM C parameter. In general, a bigger C encourages it to fit the training
# data better but might lead to overfitting. You must find the best C value
# empirically by checking how well the trained detector works on a test set of
# images you haven't trained on. Don't just leave the value set at 5. Try a
# few different C values and see what works best for your data.
options.C = 5
# Tell the code how many CPU cores your computer has for the fastest training.
options.num_threads = 4
options.be_verbose = True
training_xml_path = os.path.join(faces_folder, xml_file)
# This function does the actual training. It will save the final detector to
# detector.svm. The input is an XML file that lists the images in the training
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
dlib.train_simple_object_detector(training_xml_path, "detector.svm", options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, "detector.svm")))
# file. But for this example, we just use the training.xml file included with
# dlib.
if not os.path.exists(os.path.join(ip_folder, "detector.svm")):
os.chdir(os.path.dirname(training_xml_path))
dlib.train_simple_object_detector("training.xml",
os.path.join(ip_folder, "detector.svm"),
options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
os.chdir(os.path.dirname(training_xml_path))
train_test = dlib.test_simple_object_detector(
"training.xml",
os.path.join(ip_folder, "detector.svm").replace("\\", "/"))
print("Training accuracy: {}".format(train_test))
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
#print("Testing accuracy: {}".format(dlib.test_simple_object_detector(testing_xml_path, os.path.join(current_dir,"detector.svm")))
os.chdir(os.path.dirname(testing_xml_path))
true_test = dlib.test_simple_object_detector(
"testing.xml",
os.path.join(ip_folder, "detector.svm").replace("\\", "/"))
print("Testing accuracy: {}".format(true_test))
os.chdir(current_dir)
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
dlib.train_simple_object_detector(training_xml_path, "dog_detector.svm", options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, "dog_detector.svm")))
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
print("Testing accuracy: {}".format(
dlib.test_simple_object_detector(testing_xml_path, "dog_detector.svm")))
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
detector = dlib.simple_object_detector("dog_detector.svm")
# We can look at the HOG filter we learned. It should look like a face. Neat!
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
dlib.train_simple_object_detector(training_xml_path,
"../biblioteca/detector_focinhos.svm",
options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path,
"../biblioteca/detector_focinhos.svm")))
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
print("Testing accuracy: {}".format(
dlib.test_simple_object_detector(testing_xml_path,
"../biblioteca/detector_focinhos.svm")))
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
detector = dlib.simple_object_detector("../biblioteca/detector_focinhos.svm")
# We can look at the HOG filter we learned. It should look like a face. Neat!
win_det = dlib.image_window()
win_det.set_image(detector)
import cv2
import dlib
import os
from skimage import filter, data, io
from skimage.viewer import ImageViewer
learning_folder = 'learning'
options = dlib.simple_object_detector_training_options()
options.add_left_right_image_flips = True
options.C = 5
options.num_threads = 1
options.be_verbose = True
training_xml_path = os.path.join(learning_folder,'info.xml')
dlib.train_simple_object_detector(training_xml_path, 'info.svm', options)
print ("")
print ("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, "info.svm")
))
detector = dlib.simple_object_detector("info.svm")
win_det = dlib.image_window()
win_det.set_image(detector)
dlib.hit_enter_to_continue()
# This function does the actual training. It will save the final detector to
# detector.svm. The input is an XML file that lists the images in the training
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
dlib.train_simple_object_detector(faces_folder + "/training.xml",
"detector.svm", options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print the precision, recall, and then
# average precision.
print "\ntraining accuracy:", dlib.test_simple_object_detector(
faces_folder + "/training.xml", "detector.svm")
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
print "testing accuracy: ", dlib.test_simple_object_detector(
faces_folder + "/testing.xml", "detector.svm")
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
detector = dlib.simple_object_detector("detector.svm")
# We can look at the HOG filter we learned. It should look like a face. Neat!
win_det = dlib.image_window()
win_det.set_image(detector)
# Now let's run the detector over the images in the faces folder and display the
test_images = []
for train_image_file in train_image_files:
train_images.append(io.imread(image_floder + train_image_file))
for test_image_file in test_image_files:
test_images.append(io.imread(image_floder + test_image_file))
train_boxes = [([dlib.rectangle(left=371, top=455, right=723, bottom=943)]),
([dlib.rectangle(left=357, top=471, right=357+369, bottom=471+555)]),
([dlib.rectangle(left=413, top=313, right=413+319, bottom=313+451)]),
([dlib.rectangle(left=411, top=399, right=411+351, bottom=399+507)]),
([dlib.rectangle(left=287, top=439, right=287+423, bottom=439+537)]),
([dlib.rectangle(left=353, top=427, right=353+307, bottom=427+477)]),
([dlib.rectangle(left=369, top=569, right=369+385, bottom=569+601)]),
([dlib.rectangle(left=401, top=449, right=401+301, bottom=449+451)]),
([dlib.rectangle(left=265, top=577, right=265+417, bottom=577+597)]),
([dlib.rectangle(left=335, top=311, right=335+397, bottom=311+617)]),
([dlib.rectangle(left=341, top=381, right=341+447, bottom=381+645)]),
([dlib.rectangle(left=217, top=533, right=217+449, bottom=533+653)])]
test_boxes = [([dlib.rectangle(left=367, top=427, right=367+305, bottom=427+459)]),
([dlib.rectangle(left=481, top=309, right=481+473, bottom=309+709)]),
([dlib.rectangle(left=261, top=393, right=261+537, bottom=393+777)])]
detector = dlib.train_simple_object_detector(train_images, train_boxes, options)
# 保存模型
detector.save('stools.model')
print("\nTraining accuracy: {}".format(
dlib.test_simple_object_detector(train_images, train_boxes, detector)))
print(dlib.test_simple_object_detector(test_images, test_boxes, detector))
# This function does the actual training. It will save the final detector to
# detector.svm. The input is an XML file that lists the images in the training
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
dlib.train_simple_object_detector(faces_folder+"/training.xml","detector.svm", options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print the precision, recall, and then
# average precision.
print "\ntraining accuracy:", dlib.test_simple_object_detector(faces_folder+"/training.xml", "detector.svm")
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
print "testing accuracy: ", dlib.test_simple_object_detector(faces_folder+"/testing.xml", "detector.svm")
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
detector = dlib.simple_object_detector("detector.svm")
# We can look at the HOG filter we learned. It should look like a face. Neat!
win_det = dlib.image_window()
win_det.set_image(detector)
import os
import dlib
import cv2
import glob
print(
dlib.test_simple_object_detector("recursos/teste_relogios.xml",
"recursos/detector_relogios.svm"))
detectorRelogio = dlib.simple_object_detector("recursos/detector_relogios.svm")
for imagem in glob.glob(os.path.join("relogios_teste", "*.jpg")):
img = cv2.imread(imagem)
objetosDetectados = detectorRelogio(img, 2)
for d in objetosDetectados:
e, t, d, b = (int(d.left()), int(d.top()), int(d.right()),
int(d.bottom()))
cv2.rectangle(img, (e, t), (d, b), (0, 0, 255), 2)
cv2.imshow("Detector de relogios", img)
cv2.waitKey(0)
cv2.destroyAllWindows()
def train(train, model, test, flips, C, threads, verbose):
# Now let's do the training. The train_simple_object_detector() function has a
# bunch of options, all of which come with reasonable default values. The next
# few lines goes over some of these options.
options = dlib.simple_object_detector_training_options()
#**********************REMOVE!!!!!!!!!!!!!!!!!******************
#Since faces are left/right symmetric we can tell the trainer to train a
# symmetric detector. This helps it get the most value out of the training
# data.
options.add_left_right_image_flips = flips
# The trainer is a kind of support vector machine and therefore has the usual
# SVM C parameter. In general, a bigger C encourages it to fit the training
# data better but might lead to overfitting. You must find the best C value
# empirically by checking how well the trained detector works on a test set of
# images you haven't trained on. Don't just leave the value set at 5. Try a
# few different C values and see what works best for your data.
options.C = C
# Tell the code how many CPU cores your computer has for the fastest training.
options.num_threads = threads
options.be_verbose = verbose
# You just need to put your images into a list. TrainingImages takes a folder
#and extracts the images and bounding boxes for each image
trainingSet = TrainingImages(train)
images = trainingSet.images
# Then for each image you make a list of rectangles which give the pixel
# locations of the edges of the boxes.
# And then you aggregate those lists of boxes into one big list and then call
# train_simple_object_detector().
boxes = trainingSet.boxes
testImages = images
testBoxes = boxes
if test != train:
testSet = TrainingImages(test)
testImages = testSet.images
testBoxes = testSet.boxes
count = 1
width = 0
height = 0
##Calculating boxes, aspect ratios etc. ***May need to adjust logic
##to accommodate ambiguity.
##Also saving new masked images to disk to verify the correct
##annotations are being detected.
if not os.path.exists(train + "/masked"):
os.makedirs(train + "/masked")
aspRatios = []
flatARs = []
dictARs = {}
for j, i in enumerate(boxes):
curImageName = trainingSet.imageNames[j]
print "Image: ", curImageName
newImage = images[j].copy()
aRs = []
for box in i:
cv2.rectangle(newImage, (box.top(), box.left()),
(box.bottom(), box.right()), (255, 255, 255),
thickness=-3)
width += box.width()
height += box.height()
ar = float(box.width()) / float(box.height())
aRs.append(ar)
dictARs[ar] = box
flatARs.append(ar)
count += 1
print "Box: ", box, "\t Area: ", box.area(), "\tAR: ", aRs
aspRatios.append(aRs)
print "\nAspect Ratios: ", aspRatios, "\nDictionary: ", dictARs
baseName = curImageName.split("/")[-1]
newImageName = train + "/masked/" + (baseName).replace(
".jpg", "-boxes.jpg")
print "\nSaving: ", newImageName, "\n"
cv2.imwrite(newImageName, newImage)
##Calculating the mean and standard deviation (May not need mean
##not currently using it...)
aRMean = np.mean(flatARs, 0)
aRStd = np.std(flatARs, 0)
print "Aspect Ratio Mean: ", aRMean, " Std: ", aRStd
target_size = float(width / count) * float(height / count)
#Update the sliding window size based on input data
width, height = PlateDetector.bestWindow(boxes,
target_size=target_size)
targetSize = int(width * height)
targetAr = float(width) / height
options.detection_window_size = targetSize
print "New Width: ", width, "\tNew Height", height, "!!!"
print "Target size: ", targetSize, " Target AR: ", targetAr
##Deleting boxes with aspect ratios that are above or below the target
##aspect ratio plus one standard deviation. bestWindow estimates a target
##aspect ratio close to (but not exactly) the mean. This logic was borrowed
##from a dlib C++ HOG training example. Not sure why they didn't just use the
##mean, but this seems to work fine.
for i, imgArs in enumerate(aspRatios):
for boxArs in imgArs:
if (boxArs > (targetAr + aRStd)) or (boxArs <
(targetAr - aRStd)):
print "Deleting box ", dictARs[boxArs]
boxes[i].remove(dictARs[boxArs])
print "New boxes: ", boxes
#Train
detector = dlib.train_simple_object_detector(images, boxes, options)
# We could save this detector to disk by uncommenting the following.
detector.save(model)
# Now let's look at its HOG filter!
# win_det.set_image(detector)
# dlib.hit_enter_to_continue()
win_det = dlib.image_window()
win_det.set_image(detector)
# Note that you don't have to use the XML based input to
# test_simple_object_detector(). If you have already loaded your training
# images and bounding boxes for the objects then you can call it as shown
# below.
print("\nTraining accuracy: {}".format(
dlib.test_simple_object_detector(testImages, testBoxes, detector)))
# This function does the actual training. It will save the final detector to
# detector.svm. The input is an XML file that lists the images in the training
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
dlib.train_simple_object_detector(training_xml_path, "detector.svm", options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, "detector.svm")))
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
print("Testing accuracy: {}".format(
dlib.test_simple_object_detector(testing_xml_path, "detector.svm")))
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
detector = dlib.simple_object_detector("detector.svm")
# We can look at the HOG filter we learned. It should look like a face. Neat!
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
print "start training"
dlib.train_simple_object_detector(training_xml_path, "detector.svm", options)
print "end training"
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print ("") # Print blank line to create gap from previous output
print ("Training accuracy: {}".format(dlib.test_simple_object_detector(training_xml_path, "detector.svm")))
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
print ("Testing accuracy: {}".format(dlib.test_simple_object_detector(testing_xml_path, "detector.svm")))
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
detector = dlib.simple_object_detector("detector.svm")
# We can look at the HOG filter we learned. It should look like a face. Neat!
win_det = dlib.image_window()
win_det.set_image(detector)
# Now let's run the detector over the images in the faces folder and display the
# -*- coding: utf-8 -*-
import os
from glob import glob
import cv2
import dlib
print(
dlib.test_simple_object_detector(
'../materials/recursos/test_clock.xml',
'../materials/recursos/clock_detector.svm'))
clock_detector = dlib.simple_object_detector(
'../materials/recursos/clock_detector.svm')
for image in glob(os.path.join('../materials/relogios_teste', '*.jpg')):
img = cv2.imread(image)
detected_objects = clock_detector(img, 2)
for _object in detected_objects:
l, t, r, b = int(_object.left()), int(_object.top()), int(
_object.right()), int(_object.bottom())
cv2.rectangle(img, (l, t), (r, b), (0, 255, 255), 2)
cv2.imshow('Relógios detectados', img)
cv2.waitKey(0)
cv2.destroyAllWindows()
"""
This file demos the training of an SVM for object detection.
You should already a have labeled training set in the form of XML file.
Use imglab to create one if you have not.
"""
import dlib, os
options = dlib.simple_object_detector_training_options()
# Enable this option if your object is symmetrical.
# options.add_left_right_image_flips = True
options.C = 3 # Specify penalty parameter, too large - overfit, too small - underfit
options.num_threads = 4
options.be_verbose = True
training_xml_path = os.path.abspath('training.xml')
detector_svm_name = "detector.svm"
# Train SVM
dlib.train_simple_object_detector(training_xml_path, detector_svm_name,
options)
# Validate SVM
# Should use a different image set for testing
accuracy_result = dlib.test_simple_object_detector(training_xml_path,
detector_svm_name)
def main():
#If cleanup is True then the new images and annotations will be appended to previous ones
cleanup = True
cv2.namedWindow('frame', cv2.WINDOW_NORMAL)
cv2.resizeWindow('frame', 1920, 1080)
cv2.moveWindow("frame", 0, 0)
cap = cv2.VideoCapture(0, cv2.CAP_DSHOW)
x1, y1 = 0, 0
window_width = 190
window_height = 190
#Save images after every 4 frames to prevent duplicates
skip_frames = 4
frame_gap = 0
# Store images here
directory = 'train_images_h'
box_file = 'boxes_h.txt'
if cleanup:
# Delete the images directory if it exists
if os.path.exists(directory):
shutil.rmtree(directory)
open(box_file, 'w').close()
# Initialize the counter to 0
counter = 0
elif os.path.exists(box_file):
#Append new boxes to the previously stored ones
with open(box_file, 'r') as text_file:
box_content = text_file.read()
# Set the counter to the previous highest checkpoint
counter = int(box_content.split(':')[-2].split(',')[-1])
fr = open(box_file, 'a')
if not os.path.exists(directory):
os.mkdir(directory)
initial_wait = 0
while (True):
ret, frame = cap.read()
if not ret:
break
# Invert the image laterally to get the mirror reflection.
frame = cv2.flip(frame, 1)
# Make a copy of the original frame
orig = frame.copy()
# Wait the first 50 frames so that you can place your hand correctly
if initial_wait > 60:
frame_gap += 1
# Move the window to the right
if x1 + window_width < frame.shape[1]:
x1 += 4
time.sleep(0.1)
elif y1 + window_height + 270 < frame.shape[1]:
#Move window down by 80px
y1 += 80
x1 = 0
frame_gap = 0
initial_wait = 0
else:
break
else:
initial_wait += 1
# Save the image every nth frame.
if frame_gap == skip_frames:
# Set the image name equal to the counter value
img_name = str(counter) + '.png'
# Save the Image in the defined directory
img_full_name = directory + '/' + str(counter) + '.png'
cv2.imwrite(img_full_name, orig)
# Save bounding box coordinates
fr.write('{}:({},{},{},{}),'.format(counter, x1, y1,
x1 + window_width,
y1 + window_height))
counter += 1
frame_gap = 0
# Draw the sliding window
cv2.rectangle(frame, (x1, y1), (x1 + window_width, y1 + window_height),
(0, 255, 0), 3)
# Display the frame
cv2.imshow('frame', frame)
if cv2.waitKey(1) == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
fr.close()
print("[INFO] Data collection complete...")
# In this dictionary our images and annotations will be stored.
data = {}
# Get the indexes of all images.
image_indexes = [
int(img_name.split('.')[0]) for img_name in os.listdir(directory)
]
# Shuffle the indexes to have random train/test split later on.
np.random.shuffle(image_indexes)
# Open and read the content of the boxes.txt file
f = open(box_file, "r")
box_content = f.read()
# Convert the bounding boxes to dictionary in the format `index: (x1,y1,x2,y2)` ...
box_dict = eval('{' + box_content + '}')
# Close the file
f.close()
# Loop over all indexes
for index in image_indexes:
# Read the image in memmory and append it to the list
img = cv2.imread(os.path.join(directory, str(index) + '.png'))
# Read the associated bounding_box
bounding_box = box_dict[index]
# Convert the bounding box to dlib format
x1, y1, x2, y2 = bounding_box
dlib_box = [dlib.rectangle(left=x1, top=y1, right=x2, bottom=y2)]
# Store the image and the box together
data[index] = (img, dlib_box)
# This is the percentage of data we will use to train
# The rest will be used for testing
percent = 0.8
# How many examples make 80%.
split = int(len(data) * percent)
# Seperate the images and bounding boxes in different lists.
images = [tuple_value[0] for tuple_value in data.values()]
bounding_boxes = [tuple_value[1] for tuple_value in data.values()]
# Initialize object detector Options
options = dlib.simple_object_detector_training_options()
# I'm disabling the horizontal flipping, becauase it confuses the detector
# if you're training on few examples
options.add_left_right_image_flips = False
# Set the C parameter of SVM
options.C = 5
# Note the start time before training.
st = time.time()
# You can start the training now
print("[INFO]Beginning training of the model...")
detector = dlib.train_simple_object_detector(images[:split],
bounding_boxes[:split],
options)
# Print the Total time taken to train the detector
print('Training Completed, Total Time taken: {:.2f} seconds'.format(
time.time() - st))
file_name = 'Hand_Detector.svm'
detector.save(file_name)
win_det = dlib.image_window()
win_det.set_image(detector)
print("\nTraining Metrics: {}".format(
dlib.test_simple_object_detector(images[:split],
bounding_boxes[:split], detector)))
print("Testing Metrics: {}".format(
dlib.test_simple_object_detector(images[split:],
bounding_boxes[split:], detector)))
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
dlib.train_simple_object_detector(training_xml_path, "detector.svm", options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(
dlib.test_simple_object_detector(training_xml_path, "detector.svm")))
# However, to get an idea if it really worked without overfitting we need to
# run it on images it wasn't trained on. The next line does this. Happily, we
# see that the object detector works perfectly on the testing images.
print("Testing accuracy: {}".format(
dlib.test_simple_object_detector(testing_xml_path, "detector.svm")))
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
detector = dlib.simple_object_detector("detector.svm")
# We can look at the HOG filter we learned. It should look like a face. Neat!
plt.imshow(image[:, :, ::-1]);
plt.axis('off');
percent = 0.8
split = int(len(data) * percent)
images = [tuple_value[0] for tuple_value in data.values()]
bounding_boxes = [tuple_value[1] for tuple_value in data.values()]
options = dlib.simple_object_detector_training_options()
options.add_left_right_image_flips = False
options.C = 8
st = time.time()
#detector = dlib.train_simple_object_detector(images[:split], bounding_boxes[:split], options)
detector = dlib.train_simple_object_detector(images, bounding_boxes, options)
print('Training Completed, Total Time taken: {:.2f} seconds'.format(time.time() - st))
#file_name = 'models/Hand_Detector_v8_c8.svm'
#detector.save(file_name)
win_det = dlib.image_window()
win_det.set_image(detector)
print("Training Metrics: {}".format(dlib.test_simple_object_detector(images[:split], bounding_boxes[:split], detector)))
print("Testing Metrics: {}".format(dlib.test_simple_object_detector(images[split:], bounding_boxes[split:], detector)))
porter = Porter(detector, language='C')
output = porter.export(embed_data=True)
print(output)
# Tell the code how many CPU cores your computer has for the fastest training.
options.num_threads = 8
options.be_verbose = True
training_xml_path = "signs.xml"
## testing_xml_path = os.path.join(faces_folder, "testing.xml")
# This function does the actual training. It will save the final detector to
# detector.svm. The input is an XML file that lists the images in the training
# dataset and also contains the positions of the face boxes. To create your
# own XML files you can use the imglab tool which can be found in the
# tools/imglab folder. It is a simple graphical tool for labeling objects in
# images with boxes. To see how to use it read the tools/imglab/README.txt
# file. But for this example, we just use the training.xml file included with
# dlib.
dlib.train_simple_object_detector(training_xml_path, TRAINING, options)
# Now that we have a face detector we can test it. The first statement tests
# it on the training data. It will print(the precision, recall, and then)
# average precision.
print("") # Print blank line to create gap from previous output
print("Training accuracy: {}".format(dlib.test_simple_object_detector(training_xml_path, TRAINING)))
# Now let's use the detector as you would in a normal application. First we
# will load it from disk.
detector = dlib.simple_object_detector(TRAINING)
# We can look at the HOG filter we learned. It should look like a face. Neat!
win_det = dlib.image_window()
win_det.set_image(detector)