def VTEM_CS_IAF(videofile,
output_filename,
Wt,
Wx=2 * np.pi * 4,
Wy=2 * np.pi * 4,
start_frame=0,
end_frame=None,
dx=1.0 / 16,
dy=1.0 / 16,
fps=100,
domain=None,
h5input=True):
"""
Encode a video with IAF neurons and Centre Surround receptive field
with default parameters
Parameters
-----------
videofile : string
::
Either
1. Filename of a file containing the input video
must be stored using write_memory_to_file in simpleio.py
or using h5write in matlab
Video array with shape (a,b,c)
a: total number of frames
b: number of pixels in y direction
c: number of pixels in x direction
c is the leading dimension
in matlab should be (c,b,a)
2. Or filename of a video file if h5input is set to False.
Will throw an error if OpenCV and the required codec are
not installed
output_filename : string
output filename that will contain the spike info
Wt : float
bandwidth in t variable
if not specified, will use the info in spikefile
Wx : float, optional
bandwidth in x variable
if not specified, will use the info in spikefile
Wy : float, optional
bandwidth in y variable
if not specified, will use the info in spikefile
start_frame : integer, optional
starting frame to be encoded in the video
end_frame : integer, optional
ending frame to be encoded
if not specified, will encoding to the end of the video
dx : integer, optional
spatial resolution in x direction, distance between two pixels
dy: integer, optional
spatial resolution in y direction, distance between two pixels
fps : integer, optional
frames per second of the video
domain : list, optional
list of 2, [a,b], specifying the domain to encode
a in x direction, b in y direction
will only encode the center of the video with size [a,b]
if not specified, the whole video screen will be encoded.
h5input : bool, optional
True if the file specified is an h5 file.
False if the file specified is a video file.
If not specified, is set to True
Notes
-----
The coordinate system is given by the following
::
Row (width / X) major
-----------------------------------------> width
| Y
| ^-------------------->
| |-------------------->
| |-------------------->
| |--------------------> X
|
|
v
height
To specify receptive field parameters, create an object using
vrf.vrf_cs() with desired parameters and use function
VTEM_IAF
Examples
--------
>>> import atexit
>>> import pycuda.driver as cuda
>>> import numpy as np
>>> from vtem import vtem,vtdm
>>> cuda.init()
>>> context1 = cuda.Device(0).make_context()
>>> atexit.register(cuda.Context.pop)
>>> vtem.VTEM_CS_IAF('video.h5', 'spikes.h5', 2*np.pi*10)
"""
if h5input:
h5file = tables.openFile(videofile, 'r')
videoarray = h5file.root.real
else:
videoarray = vio.read_video(videofile)
_, Py, Px = videoarray.shape
cs = vrf.vrf_cs((Py, Px), domain=domain, dx=dx, dy=dy)
cs.load_parameters()
VTEM_IAF(videoarray,
output_filename,
Wt,
cs,
Wx,
Wy,
start_frame,
end_frame,
dx,
dy,
fps,
domain,
h5input,
npinput=True)
if h5input:
h5file.close()
def VTEM_CS_IAF(videofile, output_filename, Wt, Wx=2*np.pi*4, Wy=2*np.pi*4,
start_frame=0, end_frame=None, dx=1.0/16, dy=1.0/16,
fps=100, domain=None, h5input=True):
"""
Encode a video with IAF neurons and Centre Surround receptive field
with default parameters
Parameters
-----------
videofile : string
::
Either
1. Filename of a file containing the input video
must be stored using write_memory_to_file in simpleio.py
or using h5write in matlab
Video array with shape (a,b,c)
a: total number of frames
b: number of pixels in y direction
c: number of pixels in x direction
c is the leading dimension
in matlab should be (c,b,a)
2. Or filename of a video file if h5input is set to False.
Will throw an error if OpenCV and the required codec are
not installed
output_filename : string
output filename that will contain the spike info
Wt : float
bandwidth in t variable
if not specified, will use the info in spikefile
Wx : float, optional
bandwidth in x variable
if not specified, will use the info in spikefile
Wy : float, optional
bandwidth in y variable
if not specified, will use the info in spikefile
start_frame : integer, optional
starting frame to be encoded in the video
end_frame : integer, optional
ending frame to be encoded
if not specified, will encoding to the end of the video
dx : integer, optional
spatial resolution in x direction, distance between two pixels
dy: integer, optional
spatial resolution in y direction, distance between two pixels
fps : integer, optional
frames per second of the video
domain : list, optional
list of 2, [a,b], specifying the domain to encode
a in x direction, b in y direction
will only encode the center of the video with size [a,b]
if not specified, the whole video screen will be encoded.
h5input : bool, optional
True if the file specified is an h5 file.
False if the file specified is a video file.
If not specified, is set to True
Notes
-----
The coordinate system is given by the following
::
Row (width / X) major
-----------------------------------------> width
| Y
| ^-------------------->
| |-------------------->
| |-------------------->
| |--------------------> X
|
|
v
height
To specify receptive field parameters, create an object using
vrf.vrf_cs() with desired parameters and use function
VTEM_IAF
Examples
--------
>>> import atexit
>>> import pycuda.driver as cuda
>>> import numpy as np
>>> from vtem import vtem,vtdm
>>> cuda.init()
>>> context1 = cuda.Device(0).make_context()
>>> atexit.register(cuda.Context.pop)
>>> vtem.VTEM_CS_IAF('video.h5', 'spikes.h5', 2*np.pi*10)
"""
if h5input:
h5file = tables.openFile(videofile, 'r')
videoarray = h5file.root.real
else:
videoarray = vio.read_video(videofile)
_ , Py, Px = videoarray.shape
cs = vrf.vrf_cs((Py,Px), domain=domain, dx=dx, dy=dy)
cs.load_parameters()
VTEM_IAF(videoarray, output_filename, Wt, cs, Wx, Wy,start_frame,
end_frame, dx, dy, fps, domain, h5input, npinput=True)
if h5input:
h5file.close()
def VTEM_IAF(videofile,
output_filename,
Wt,
rf,
Wx=2 * np.pi * 4,
Wy=2 * np.pi * 4,
start_frame=0,
end_frame=None,
dx=1.0 / 16,
dy=1.0 / 16,
fps=100,
domain=None,
h5input=True,
npinput=False):
"""
Encode a video with IAF neurons and the specified Receptive field
Parameters
-----------
videofile : string
::
Either
1. Filename of a file containing the input video
must be stored using write_memory_to_file in simpleio.py
or using h5write in matlab
Video array with shape (a,b,c)
a: total number of frames
b: number of pixels in y direction
c: number of pixels in x direction
c is the leading dimension
in matlab should be (c,b,a)
2. Or filename of a video file if h5input is set to False.
Will throw an error if OpenCV and the required codec are
not installed
output_filename : string
output filename that will contain the spike info
Wt : float
bandwidth in t variable
if not specified, will use the info in spikefile
rf : vrf
Receptive field object.
Wx : float, optional
bandwidth in x variable
if not specified, will use the info in spikefile
Wy : float, optional
bandwidth in y variable
if not specified, will use the info in spikefile
start_frame : integer, optional
starting frame to be encoded in the video
end_frame : integer, optional
ending frame to be encoded
if not specified, will encoding to the end of the video
dx : integer, optional
spatial resolution in x direction, distance between two pixels
dy : integer, optional
spatial resolution in y direction, distance between two pixels
fps : integer, optional
frames per second of the video
domain : list, optional
list of 2, [a,b], specifying the domain to encode
a in x direction, b in y direction
will only encode the center of the video with size [a,b]
if not specified, the whole video screen will be encoded.
h5input : bool, optional
True if the file specified is an h5 file.
False if the file specified is a video file.
If not specified, is set to True
npinput : bool, optional
True if videofile is a numpy array.
If not specified, will be set to False.
Notes
-----
The coordinate system is given by the following
::
Row (width / X) major
-----------------------------------------> width
| Y
| ^-------------------->
| |-------------------->
| |-------------------->
| |--------------------> X
|
|
v
height
Examples
--------
>>> import atexit
>>> import pycuda.driver as cuda
>>> import numpy as np
>>> from vtem import vtem,vtdm
>>> cuda.init()
>>> context1 = cuda.Device(0).make_context()
>>> atexit.register(cuda.Context.pop)
>>> gb = vrf.vrf_gabor((Py,Px), domain=domain, dx=dx, dy=dy)
>>> gb.load_parameters()
>>> gb.get_gabor_parameters(dilations = [1, 0, -1, -2, -3], num_rotations = 8,
alpha0 = np.float64(2), b0 = np.float64(1.5),
b0scalebase=np.sqrt(2), KAPPA = 3)
>>> gb.load_parameters(set=1)
>>> vtem.VTEM_Gabor_IAF('video.h5', 'spikes.h5', 2*np.pi*10, gb)
"""
if npinput:
videoarray = videofile
h5input = False
else:
if h5input:
h5file = tables.openFile(videofile, 'r')
videoarray = h5file.root.real
else:
videoarray = vio.read_video(videofile)
TotalFrames, Py, Px = videoarray.shape
if end_frame is None:
end_frame = TotalFrames
if end_frame > TotalFrames:
end_frame = TotalFrames
print "warning: the end frame is changed to %d, which " \
"is the maximum frame in the video file\n" % end_frame
dt = 1.0 / fps
T = (end_frame - start_frame) * dt
dx = float(dx)
dy = float(dy)
print "number of neurons: %d" % (rf.num_neurons)
if rf.__class__.__name__ == 'vrf_gabor':
rftype = 'gabor'
elif rf.__class__.__name__ == 'vrf_cs':
rftype = 'cs'
else:
raise TypeError('Receptive field type not recognized\n.')
spikefile = ss.vtem_storage(output_filename, rf.num_neurons, rftype)
spikefile.write_video_attributes(Wt, Wx, Wy, rf.Px, rf.Py, dx, dy)
if rftype == 'gabor':
spikefile.write_gabor_parameters(rf)
elif rftype == 'cs':
spikefile.write_cs_parameters(rf)
iaf = ee.IAF_encode(rf.num_neurons, dt)
iaf.load_parameters()
spikefile.write_neuron_parameters(iaf)
freemem, totalmem = cuda.mem_get_info()
ONE_TIME_FRAMES = min(101, np.int(min(end_frame - start_frame + 1, \
2**(np.floor(np.log2((freemem/5) / videoarray.dtype.itemsize / \
Px / Py))))))
spikefile.write_refresh_interval((ONE_TIME_FRAMES - 1) * dt)
print "encoding..."
total_spikes = 0
block_video = 0
for i in range(start_frame, end_frame, ONE_TIME_FRAMES - 1):
if (h5input):
block_video = videoarray.read(i, i + ONE_TIME_FRAMES)
else:
block_video = videoarray[i:i + ONE_TIME_FRAMES, :, :]
output = rf.filter(block_video)
spikes, spike_count = iaf.encode(output, avg_rate=0.1)
del output
cum_spike_count = np.concatenate(
(np.zeros(1, np.int32), np.cumsum(spike_count)))
all_spikes = np.empty(cum_spike_count[-1])
for j in range(rf.num_neurons):
all_spikes[cum_spike_count[j]:cum_spike_count[j + 1]] = spikes[j]
spikefile.write_spike_in_timebin(spike_count, all_spikes)
del all_spikes
iaf.reset_timer()
total_spikes += cum_spike_count[-1]
print "%.0f%%" % (float((i + ONE_TIME_FRAMES-1 - start_frame)) / \
(end_frame - start_frame) *100)
spikefile.close()
if h5input:
h5file.close()
print "done, total spikes %d" % (total_spikes)
def VTEM_IAF(videofile, output_filename, Wt, rf, Wx=2*np.pi*4, Wy=2*np.pi*4,
start_frame=0, end_frame=None, dx=1.0/16, dy=1.0/16,
fps=100, domain=None, h5input=True, npinput=False):
"""
Encode a video with IAF neurons and the specified Receptive field
Parameters
-----------
videofile : string
::
Either
1. Filename of a file containing the input video
must be stored using write_memory_to_file in simpleio.py
or using h5write in matlab
Video array with shape (a,b,c)
a: total number of frames
b: number of pixels in y direction
c: number of pixels in x direction
c is the leading dimension
in matlab should be (c,b,a)
2. Or filename of a video file if h5input is set to False.
Will throw an error if OpenCV and the required codec are
not installed
output_filename : string
output filename that will contain the spike info
Wt : float
bandwidth in t variable
if not specified, will use the info in spikefile
rf : vrf
Receptive field object.
Wx : float, optional
bandwidth in x variable
if not specified, will use the info in spikefile
Wy : float, optional
bandwidth in y variable
if not specified, will use the info in spikefile
start_frame : integer, optional
starting frame to be encoded in the video
end_frame : integer, optional
ending frame to be encoded
if not specified, will encoding to the end of the video
dx : integer, optional
spatial resolution in x direction, distance between two pixels
dy : integer, optional
spatial resolution in y direction, distance between two pixels
fps : integer, optional
frames per second of the video
domain : list, optional
list of 2, [a,b], specifying the domain to encode
a in x direction, b in y direction
will only encode the center of the video with size [a,b]
if not specified, the whole video screen will be encoded.
h5input : bool, optional
True if the file specified is an h5 file.
False if the file specified is a video file.
If not specified, is set to True
npinput : bool, optional
True if videofile is a numpy array.
If not specified, will be set to False.
Notes
-----
The coordinate system is given by the following
::
Row (width / X) major
-----------------------------------------> width
| Y
| ^-------------------->
| |-------------------->
| |-------------------->
| |--------------------> X
|
|
v
height
Examples
--------
>>> import atexit
>>> import pycuda.driver as cuda
>>> import numpy as np
>>> from vtem import vtem,vtdm
>>> cuda.init()
>>> context1 = cuda.Device(0).make_context()
>>> atexit.register(cuda.Context.pop)
>>> gb = vrf.vrf_gabor((Py,Px), domain=domain, dx=dx, dy=dy)
>>> gb.load_parameters()
>>> gb.get_gabor_parameters(dilations = [1, 0, -1, -2, -3], num_rotations = 8,
alpha0 = np.float64(2), b0 = np.float64(1.5),
b0scalebase=np.sqrt(2), KAPPA = 3)
>>> gb.load_parameters(set=1)
>>> vtem.VTEM_Gabor_IAF('video.h5', 'spikes.h5', 2*np.pi*10, gb)
"""
if npinput:
videoarray = videofile
h5input = False
else:
if h5input:
h5file = tables.openFile(videofile, 'r')
videoarray = h5file.root.real
else:
videoarray = vio.read_video(videofile)
TotalFrames, Py, Px = videoarray.shape
if end_frame is None:
end_frame = TotalFrames
if end_frame > TotalFrames:
end_frame = TotalFrames
print "warning: the end frame is changed to %d, which " \
"is the maximum frame in the video file\n" % end_frame
dt = 1.0 / fps
T = (end_frame - start_frame)*dt
dx = float(dx)
dy = float(dy)
print "number of neurons: %d" % (rf.num_neurons)
if rf.__class__.__name__ == 'vrf_gabor':
rftype = 'gabor'
elif rf.__class__.__name__ == 'vrf_cs':
rftype = 'cs'
else:
raise TypeError('Receptive field type not recognized\n.')
spikefile = ss.vtem_storage(output_filename, rf.num_neurons, rftype)
spikefile.write_video_attributes(Wt, Wx, Wy, rf.Px, rf.Py, dx, dy)
if rftype == 'gabor':
spikefile.write_gabor_parameters(rf)
elif rftype == 'cs':
spikefile.write_cs_parameters(rf)
iaf = ee.IAF_encode(rf.num_neurons,dt)
iaf.load_parameters()
spikefile.write_neuron_parameters(iaf)
freemem, totalmem = cuda.mem_get_info()
ONE_TIME_FRAMES = min(101, np.int(min(end_frame - start_frame + 1, \
2**(np.floor(np.log2((freemem/5) / videoarray.dtype.itemsize / \
Px / Py))))))
spikefile.write_refresh_interval((ONE_TIME_FRAMES-1)*dt)
print "encoding..."
total_spikes = 0
block_video = 0
for i in range(start_frame, end_frame, ONE_TIME_FRAMES-1):
if(h5input):
block_video = videoarray.read(i, i+ONE_TIME_FRAMES)
else:
block_video = videoarray[i:i+ONE_TIME_FRAMES,:,:]
output = rf.filter(block_video)
spikes,spike_count=iaf.encode(output, avg_rate=0.1)
del output
cum_spike_count = np.concatenate((np.zeros(1,np.int32),
np.cumsum(spike_count)))
all_spikes = np.empty(cum_spike_count[-1])
for j in range(rf.num_neurons):
all_spikes[cum_spike_count[j]:cum_spike_count[j+1]] = spikes[j]
spikefile.write_spike_in_timebin(spike_count, all_spikes)
del all_spikes
iaf.reset_timer()
total_spikes += cum_spike_count[-1]
print "%.0f%%" % (float((i + ONE_TIME_FRAMES-1 - start_frame)) / \
(end_frame - start_frame) *100)
spikefile.close()
if h5input:
h5file.close()
print "done, total spikes %d" % (total_spikes)