def deepdream_case1(image_bytes):
end = 'inception_4b/output'
# input image
img = np.float32(imread.imread_from_blob(image_bytes))
if img.shape[2] == 1:
img = skimage.color.gray2rgb(img[:,:,0])
# best match
guide = find_guide(img)
h, w = guide.shape[:2]
src, dst = net.blobs['data'], net.blobs[end]
src.reshape(1,3,h,w)
src.data[0] = preprocess(net, guide)
net.forward(end=end)
guide_features = dst.data[0].copy()
def objective_guide(dst):
x = dst.data[0].copy()
y = guide_features
ch = x.shape[0]
x = x.reshape(ch,-1)
y = y.reshape(ch,-1)
A = x.T.dot(y) # compute the matrix of dot-products with guide features
dst.diff[0].reshape(ch,-1)[:] = y[:,A.argmax(1)] # select ones that match best
result = deepdream(net, img, end=end, objective=objective_guide)
dst_image = PIL.Image.fromarray(result.astype(np.uint8))
output_stream = StringIO.StringIO()
dst_image.save(output_stream, format="jpeg")
contents = output_stream.getvalue()
output_stream.close()
base_coded = base64.b64encode(contents)
return base_coded
def plot_chunk(chunk,
mode='spectrogram',
output_folder=None,
size=227,
nfft=None,
**kwargs):
"""
Plot spectrograms for a chunk of a wav-file using the described parameters.
:param chunk: audio chunk to be plotted.
:param mode: type of audio plot to create.
:param nfft: number of samples for the fast fourier transformation (Default: 256)
:param size: size of the spectrogram plot in pixels. Height and width are always identical (Default: 227)
:param output_folder: if given, the plot is saved to this path in .png format (Default: None)
:param kwargs: keyword args for plotting functions
:return: blob of the spectrogram plot
"""
filename, sr, ts, audio = chunk
write_index = ts is not None
if not nfft:
nfft = _next_power_of_two(int(sr * 0.025))
fig = plt.figure(frameon=False)
fig.set_size_inches(1, 1)
ax = plt.Axes(
fig,
[0., 0., 1., 1.],
)
ax.set_axis_off()
fig.add_axes(ax)
with warnings.catch_warnings():
warnings.simplefilter('ignore')
spectrogram_axes = PLOTTING_FUNCTIONS[mode](audio, sr, nfft, **kwargs)
del audio
fig.add_axes(spectrogram_axes, id='spectrogram')
if output_folder:
file_name = basename(filename)[:-4]
outfile = join(
output_folder,
'{}_{:.4f}'.format(file_name, ts).rstrip('0').rstrip('.') +
'.png') if write_index else join(output_folder, file_name + '.png')
fig.savefig(outfile, format='png', dpi=size)
buf = io.BytesIO()
fig.savefig(buf, format='png', dpi=size)
buf.seek(0)
fig.clf()
plt.close(fig)
img_blob = buf.read()
buf.close()
try:
img = imread_from_blob(img_blob, 'png')
img = img[:, :, :-1]
except IOError:
print('Error while reading the spectrogram blob.')
return None
return PlotTuple(name=filename, timestamp=ts, plot=img)
def imread(record, dtype=None):
"""Load an image from a WARC record.
Parameters
----------
record : WARC Record
"""
im = _imread.imread_from_blob(record.payload.read())
if dtype is not None:
im = convert(im, dtype)
return im
def download_and_resize(url, filename, output_dir="", w_h_shape=None):
try:
req = urlopen(url)
file_type = req.headers.get('Content-Type').split('/')[1]
image = imread_from_blob(req.read(), file_type)
if w_h_shape:
image = cv2.resize(image, w_h_shape)
if output_dir and not os.path.exists(output_dir):
os.mkdir(output_dir)
file_path = os.path.join(output_dir, filename+"."+file_type)
imsave(file_path, image, formatstr=file_type)
height, width, _ = image.shape
return filename, file_path, None, width, height
except Exception as ex:
return filename, None, ex, -1, -1
def feedforward_bytes(image_bytes): src_float = skimage.img_as_float(imread.imread_from_blob(image_bytes)).astype(np.float32) if src_float.shape[2] == 1: src_float = skimage.color.gray2rgb(src_float[:,:,0]) prediction = net.predict([src_float]) return prediction
def find_guide(input_bytes):
#Dummy guide
img = np.float32(imread.imread_from_blob(open('/home/denis/workspace/DataScienceHack/pynet/deepdream/new.jpg','rb').read()))
if img.shape[2] == 1:
img = skimage.color.gray2rgb(img[:,:,0])
return img
def plot_chunk(chunk,
mode='spectrogram',
output_folder=None,
base_path=None,
size=227,
nfft=None,
file_type='png',
labelling=False,
**kwargs):
"""
Plot spectrograms for a chunk of a wav-file using the described parameters.
:param chunk: audio chunk to be plotted.
:param mode: type of audio plot to create.
:param nfft: number of samples for the fast fourier transformation \
(Default: 256)
:param size: size of the spectrogram plot in pixels. Height and width are \
always identical (Default: 227)
:param output_folder: if given, the plot is saved to this path in .png \
format (Default: None)
:param kwargs: keyword args for plotting functions
:return: blob of the spectrogram plot
"""
matplotlib.use('Agg')
import matplotlib.pyplot as plt
filename, sr, ts, audio = chunk
write_index = ts is not None
if not nfft:
nfft = _next_power_of_two(int(sr * 0.025))
log.debug(f'Using nfft={nfft} for the FFT.')
fig = plt.figure(frameon=False, tight_layout=False)
if labelling:
pass
else:
fig.set_size_inches(1, 1)
ax = plt.Axes(fig, [0., 0., 1., 1.])
ax.set_axis_off()
fig.add_axes(ax)
with warnings.catch_warnings():
warnings.simplefilter('ignore')
spectrogram_axes = PLOTTING_FUNCTIONS[mode](audio, sr, nfft, **kwargs)
if labelling:
original_xlim = spectrogram_axes.get_xlim()
if mode != 'chroma':
kHz_ticks = np.apply_along_axis(lambda x: x / 1000, 0,
spectrogram_axes.get_yticks())
spectrogram_axes.set_yticklabels(kHz_ticks)
spectrogram_axes.set_ylabel('Frequency [kHz]',
fontdict=label_font)
else:
spectrogram_axes.set_ylabel('Pitch Classes',
fontdict=label_font)
if labelling:
spectrogram_axes.set_xticks(spectrogram_axes.get_xticks()[::2])
spectrogram_axes.set_xlabel('Time [s]', fontdict=label_font)
spectrogram_axes.set_xlim(original_xlim)
del audio
fig.add_axes(spectrogram_axes, id='spectrogram')
if labelling:
plt.colorbar(format='%+2.1f dB')
plt.tight_layout()
if output_folder:
relative_file_name = f'{splitext(get_relative_path(filename, base_path))[0]}_{ts:g}.{file_type}' if write_index else f'{splitext(get_relative_path(filename, base_path))[0]}.{file_type}'
if base_path is None:
outfile = join(output_folder, basename(relative_file_name))
else:
outfile = join(output_folder, relative_file_name)
log.debug(f'Saving spectrogram plot to {outfile}.')
makedirs(dirname(outfile), exist_ok=True)
fig.savefig(outfile, format=file_type, dpi=size)
buf = io.BytesIO()
fig.savefig(buf, format='png', dpi=size)
buf.seek(0)
fig.clf()
plt.close(fig)
img_blob = buf.read()
buf.close()
try:
img = imread_from_blob(img_blob, 'png')
img = img[:, :, :-1]
log.debug(f'Read spectrogram plot with shape {img.shape}.')
except IOError:
log.error('Error while reading the spectrogram blob.')
return None
return PlotTuple(name=get_relative_path(filename, base_path),
timestamp=ts,
plot=img)
