def detect_audios(model, baseModel, audiosPath):
vgg_extract.parse_audio_files(audiosPath, 'tmpImages')
imagesPath = 'tmpImages'
featuresLength = np.prod(baseModel.layers[-1].output.shape[1:])
imagePaths = list(paths.list_images('tmpImages'))
classNames = target_names()
for imagePath in imagePaths:
features = []
batchImages = []
image = load_img(imagePath, target_size=(WIDTH, HEIGHT))
image = img_to_array(image)
image = np.expand_dims(image, axis=0)
image = imagenet_utils.preprocess_input(image)
batchImages.append(image)
batchImages = np.vstack(batchImages)
batchFeatures = baseModel.predict(batchImages,
batch_size=len(batchImages))
features.append(batchFeatures.reshape(len(batchImages),
featuresLength))
features = np.vstack(features)
preds = model.predict(features)
pred = preds[0]
print(imagePath + ": " + str(classNames[int(pred)]))
shutil.rmtree('tmpImages')
def detect_real_time(model, baseModel):
featuresLength = np.prod(baseModel.layers[-1].output.shape[1:])
imagePaths = ['test_sample.png']
classNames = target_names()
bs = 5
while True:
features = []
record_input()
for i in np.arange(0, len(imagePaths), bs):
batchPaths = imagePaths[i:i + bs]
batchImages = []
for (_, imagePath) in enumerate(batchPaths):
image = load_img(imagePath, target_size=(WIDTH, HEIGHT))
image = img_to_array(image)
image = np.expand_dims(image, axis=0)
image = imagenet_utils.preprocess_input(image)
batchImages.append(image)
batchImages = np.vstack(batchImages)
batchFeatures = baseModel.predict(batchImages,
batch_size=len(batchImages))
features.append(
batchFeatures.reshape(len(batchImages), featuresLength))
features = np.vstack(features)
pred = model.predict_proba(features)[0]
index = np.argmax(pred)
prob = pred[index]
print(str(classNames[index]) + " - " + str(prob))
def detect_real_time(model, baseModel, storeFile=False):
imagePaths = ['test_sample.png']
classNames = target_names()
try:
with sd.InputStream(samplerate=sample_rate,
callback=callback,
channels=2):
while run:
q.get()
melspectrogram_preprocess(data, sample_rate)
imagePath = imagePaths[0]
image = cv2.imread(imagePath)
image = aap.preprocess(image)
image = iap.preprocess(image)
image = np.expand_dims(image, axis=0) / 255.0
result = model.predict(image)
pred = result.argmax(axis=1)[0]
print(result)
prob = result[0][pred]
print(str(classNames[pred]) + " - " + str(prob))
except KeyboardInterrupt:
print('\nRecording finished')
def detect_images(model, baseModel, imagesPath):
imagePaths = list(paths.list_images(imagesPath))
classNames = target_names()
labels = [p.split(os.path.sep)[-2] for p in imagePaths]
preds = []
for i in np.arange(0, len(imagePaths)):
imagePath = imagePaths[i]
image = cv2.imread(imagePath)
image = aap.preprocess(image)
image = iap.preprocess(image)
image = np.expand_dims(image, axis=0) / 255.0
result = model.predict(image)
pred = result.argmax(axis=1)[0]
preds.append(pred)
labels = np.vstack(labels)
lb = LabelEncoder().fit(np.vstack(classNames))
labels = lb.transform(labels)
print(classification_report(labels, preds, target_names=classNames))
def detect_audios(model, baseModel, audiosPath):
vgg_extract.parse_audio_files(audiosPath, 'tmpImages')
imagesPath = 'tmpImages'
imagePaths = list(paths.list_images('tmpImages'))
classNames = target_names()
for imagePath in imagePaths:
image = cv2.imread(imagePath)
image = aap.preprocess(image)
image = iap.preprocess(image)
image = np.expand_dims(image, axis=0) / 255.0
result = model.predict(image)
pred = result.argmax(axis=1)[0]
print(imagePath + ": " + str(classNames[int(pred)]))
shutil.rmtree('tmpImages')
def detect_real_time(model, baseModel):
imagePaths = ['test_sample.png']
classNames = target_names()
while True:
record_input()
imagePath = imagePaths[0]
image = cv2.imread(imagePath)
image = aap.preprocess(image)
image = iap.preprocess(image)
image = np.expand_dims(image, axis=0) / 255.0
result = model.predict(image)
pred = result.argmax(axis=1)[0]
print(result)
prob = result[0][pred]
print(str(classNames[pred]) + " - " + str(prob))
def detect_images(model, baseModel, imagesPath):
featuresLength = np.prod(baseModel.layers[-1].output.shape[1:])
imagePaths = list(paths.list_images(imagesPath))
classNames = target_names()
bs = 5
labels = [p.split(os.path.sep)[-2] for p in imagePaths]
features = []
for i in np.arange(0, len(imagePaths), bs):
batchPaths = imagePaths[i:i + bs]
batchImages = []
for (_, imagePath) in enumerate(batchPaths):
image = load_img(imagePath, target_size=(WIDTH, HEIGHT))
image = img_to_array(image)
image = np.expand_dims(image, axis=0)
image = imagenet_utils.preprocess_input(image)
batchImages.append(image)
batchImages = np.vstack(batchImages)
batchFeatures = baseModel.predict(batchImages,
batch_size=len(batchImages))
features.append(batchFeatures.reshape(len(batchImages),
featuresLength))
labels = np.vstack(labels)
lb = LabelEncoder().fit(np.vstack(classNames))
labels = lb.transform(labels)
features = np.vstack(features)
preds = model.predict(features)
print(classification_report(labels, preds, target_names=classNames))