def putincnn(self, tensor_track, model):
# put data in cnn
print("predicting...")
print('-' * 50)
output = predict(tensor_track, model)
print(output)
return output
def upload_image():
"""
Returns JSON in the form:
{
'class': str(cls),
'probability': float(conf),
'blurredImage': str(encoded_img),
'maskImage': str(encoded_img_mask),
'unblurredMask': str(encoded_img_saliency)
}
Where images are in base64.
@return: JSON
"""
data = request.get_json()
# Handle empty request
if data is None:
print("[ ERR ] Request body invalid: empty")
return "Error: Missing Request"
# Decode image and predict output
img = Image.open(BytesIO(base64.urlsafe_b64decode(data['img'])))
print("[ INF ] Success: Image Loaded")
return jsonify(load_model.predict(img))
waveform = waveform.numpy()[0, :]
length = np.shape(waveform)[0]
for index in range(0, length, STEP_SIZE):
waveform_part = waveform[index:index + int(FRAME_SIZE)] # [44100]
if len(waveform_part) < FRAME_SIZE:
break
waveform_part = waveform_part[np.newaxis, ...] # [1, 44100]
waveform_part = torch.from_numpy(waveform_part) # torch [1, 44100]
print(np.shape(waveform_part))
mel_specgram = torchaudio.transforms.MelSpectrogram(new_sample_rate)(
waveform_part) # torch [1, 128, 221]
mel_specgram = mel_specgram.detach().numpy() # numpy [1, 128, 221]
if np.shape(waveform_part)[1] == int(FRAME_SIZE):
track_array.append(mel_specgram)
tensor_track = torch.Tensor(track_array)
# load model
model = torch.load('../model/harmonica_model/harmonica_error_2d_model_15.pth')
print("loading model...")
print('-' * 50)
# put data in cnn
print("predicting...")
print('-' * 50)
output = predict(tensor_track, model)
print(output)
x, y, w, h = 0, 0, 0, 0
while ret:
ret, framecolor = cap.read()
frame = cv2.cvtColor(framecolor, cv2.COLOR_BGR2GRAY)
frame_sub = backsub.apply(frame)
contours, hierarchy = cv2.findContours(frame_sub, cv2.RETR_EXTERNAL,
cv2.CHAIN_APPROX_NONE)
for item in contours:
if cv2.contourArea(item) > 350:
x, y, w, h = cv2.boundingRect(item)
croped = frame[y:y + h, x:x + w]
resized = cv2.resize(croped, dsize=(64, 64))
img_tensor = np.reshape(resized, newshape=(1, 64, 64, 1))
# now predict
k = predict(img_tensor)
img = cv2.rectangle(framecolor, (x, y), (x + w, y + h),
color=(0, 255, 0),
thickness=3)
if k == 1:
cv2.putText(img,
'human',
org=(x, y),
fontFace=cv2.FONT_HERSHEY_COMPLEX,
fontScale=1,
color=(255, 0, 0),
bottomLeftOrigin=False)
cv2.imshow(winname='video', mat=img)
if cv2.waitKey(1) & 0xFF == ord('q'):
break