def __init__(self):
self.inference_model = inference.Model()
self.default_metrics_dict = {
'accuracy': accuracy_score,
'f1': f1_score,
'recall': recall_score,
'precision': precision_score,
}
cnt = 0
repeat_action = False
filename = None
data = None
cnn_model = None
last_operation = None
EMPTY = np.array([])
gui = gui.GUI(interface=itfc, dataset=dataset)
if dataset.model and not args.browser:
import inference
cnn_model = inference.Model(dataset)
root = Tk.Tk()
if args.fullscreen_mode:
root.wm_title("")
else:
root.wm_title("Oscilloscope")
if args.browser:
fig, ax = plt.subplots(1, 1, figsize=(10, 4))
else:
fig, ax = plt.subplots(1, 1, figsize=(11, 4))
fig.subplots_adjust(bottom=0.15)
frame = Tk.Frame(master=root)
frame_row0 = Tk.Frame(master=frame)
import argparse
from flask import Flask, jsonify, request
import inference
__author__ = 'Junior Teudjio'
app = Flask(__name__)
inference_model = inference.Model()
@app.route('/predict', methods=['POST', 'GET'])
def predict():
request_args = request.args
prediction = inference_model.predict(image_path=request_args['image_path'],
model_type=request_args['model_type'])
return jsonify(prediction)
def _setup_args():
parser = argparse.ArgumentParser()
parser.add_argument('--port', type=int, default=8383)
parser.add_argument('--host', type=str, default='0.0.0.0')
return parser.parse_args()
if __name__ == '__main__':
args = _setup_args()
app.run(port=args.port,host=args.host)
def main():
parser, args = parse()
print('list')
if args.list_devices:
print(sounddevice.query_devices())
parser.exit(0)
if args.samplerate is None:
device_info = sounddevice.query_devices(args.device, 'input')
args.samplerate = device_info['default_samplerate']
mapping = [c - 1 for c in args.channels] # Channel numbers start with 1
q = queue.Queue()
def audio_callback(indata, frames, time, status):
"""This is called (from a separate thread) for each audio block."""
if status:
print(status, file=sys.stderr)
# Fancy indexing with mapping creates a (necessary!) copy:
q.put(indata[:, mapping])
#print(indata.shape)
stream = sounddevice.InputStream(
device=args.device, channels=max(args.channels),
samplerate=args.samplerate, callback=audio_callback)
print('stream open', args.device, stream)
assert args.samplerate == 16000
n_samples = int(args.samplerate*0.960*1.05)
hop_length = n_samples * (1-args.overlap)
new_samples = 0
audio_buffer = numpy.zeros(shape=(n_samples,))
model = yamnet.Model()
import keras
import tensorflow
with stream:
while True:
data = q.get()
data = numpy.squeeze(data)
#print(data.shape)
# move existing data over
audio_buffer = numpy.roll(audio_buffer, len(data), axis=0)
# add the new data
audio_buffer[len(audio_buffer)-len(data):len(audio_buffer)] = data
# check if we have received enough new data to do new classification
new_samples += len(data)
if new_samples >= hop_length:
t = datetime.datetime.now()
new_samples = 0
waveform = audio_buffer
with model.graph.as_default():
#x = numpy.reshape(waveform, [1, -1])
x = numpy.expand_dims(waveform, 0)
#x = numpy.expand_dims(x, -1)
scores, _spec, embeddings = model.yamnet.predict(x, steps=1, batch_size=1)
# Report the highest-scoring classes and their scores.
prediction = numpy.mean(scores, axis=0)
top_n = 3
top = numpy.argsort(prediction)[::-1][:top_n]
def format_pred(i):
return ' {:12s}: {:.3f}'.format(model.class_names[i], prediction[i])
if numpy.max(prediction) > 0.3:
print('classify', t)
print('\n'.join(format_pred(i) for i in top))
print('stopped')