def __init__(
self,
nRows,
nCols,
vol_preserve,
base,
nLevels,
valid_outside=True,
tess='tri',
sigma_lm=1000 / 10, # worked pretty well (hands data)
# sigma_lm = 1000/100, # Didn't work so well
scale_spatial=1.0 * .1,
scale_value=100,
zero_v_across_bdry=[False] * 2,
wlp=1e-4,
scale_quiver=None):
self.base = base
self.nLevels = nLevels
self.sigma_lm = sigma_lm
self.wlp = wlp
self.tw = TransformWrapper(nRows=nRows,
nCols=nCols,
vol_preserve=vol_preserve,
nLevels=nLevels,
base=base,
scale_spatial=scale_spatial,
scale_value=scale_value,
tess=tess,
valid_outside=valid_outside,
zero_v_across_bdry=zero_v_across_bdry)
if scale_quiver is None:
raise ValueError
self.scale_quiver = scale_quiver
def __init__(self,nRows=100,
nCols=100,
base = [2,2],
nLevels=4,
tess='tri',
zero_v_across_bdry=[False]*2,
# range_max_val = 1, # FOR NOW, KEEP IT.
# nPtsDense=10000,
scale_spatial=1.0 * 10,
scale_value=2.0,
sigma_signal=None,
wlp=1e-4,
ll_type=['gaussian','gaussian_on_distancetransform'][0],
only_local=False,
valid_outside=True):
ll_type = ll_type.lower()
if ll_type == 'gaussian':
self.SDLL=SDLL_gaussian
# elif ll_type =='gaussian_on_distancetransform':
# self.SDLL=SDLL_gaussian_on_distancetransform
# elif ll_type == 'vmf':
# self.SDLL=SDLL_VMF
else:
raise ValueError(ll_type)
# self.range_min_val = 0.0
# self.range_max_val = range_max_val
self.base = base
self.nLevels=nLevels
if sigma_signal is None:
raise ValueError("sigma_signal cannot be None")
self.sigma_signal = sigma_signal
self.wlp = wlp
self.tw = TransformWrapper(nRows=nRows,nCols=nCols,
nLevels=nLevels,
base=base,
tess=tess,
# nPtsDense=nPtsDense,
scale_spatial=scale_spatial,
scale_value=scale_value,
zero_v_across_bdry=zero_v_across_bdry,
only_local=only_local,
valid_outside=valid_outside
)
def example(img=None,tess='I',eval_cell_idx=True,eval_v=True,show_downsampled_pts=True,
valid_outside=True,base=[1,1],
scale_spatial=.1,
scale_value=100,
permute_cell_idx_for_display=True,
nLevels=3,
vol_preserve=False,
zero_v_across_bdry=[0,0],
use_lims_when_plotting=True):
show_downsampled_pts = bool(show_downsampled_pts)
eval_cell_idx = bool(eval_cell_idx)
eval_v = bool(eval_cell_idx)
valid_outside = bool(valid_outside)
permute_cell_idx_for_display = bool(permute_cell_idx_for_display)
vol_preserve = bool(vol_preserve)
if img is None:
img = Img(get_std_test_img())
else:
img=Img(img)
img = img[:,:,::-1] # bgr2rgb
tw = TransformWrapper(nRows=img.shape[0],
nCols=img.shape[1],
nLevels=nLevels,
base=base,
scale_spatial=scale_spatial, # controls the prior's smoothness
scale_value=scale_value, # controls the prior's variance
tess=tess,
vol_preserve=vol_preserve,
zero_v_across_bdry=zero_v_across_bdry,
valid_outside=valid_outside)
print tw
# You probably want to do that: padding image border with zeros
border_width=1
img[:border_width]=0
img[-border_width:]=0
img[:,:border_width]=0
img[:,-border_width:]=0
# The tw.calc_T_fwd (or tw.calc_T_inv) is always done in gpu.
# After using it to compute new pts,
# you may want to use remap (to warp an image accordingly).
# If you will use tw.remap_fwd (or tw.remap_inv), which is done in gpu,
# then the image type can be either float32 or float64.
# But if you plan to use tw.tw.remap_fwd_opencv (or tw.remap_inv_opencv),
# which is done in cpu (hence slightly lower) but supports better
# interpolation methods, then the image type must be np.float32.
# img_original = CpuGpuArray(img.copy().astype(np.float32))
img_original = CpuGpuArray(img.copy().astype(np.float64))
img_wrapped_fwd= CpuGpuArray.zeros_like(img_original)
img_wrapped_bwd= CpuGpuArray.zeros_like(img_original)
seed=0
np.random.seed(seed)
ms_Avees=tw.get_zeros_PA_all_levels()
ms_theta=tw.get_zeros_theta_all_levels()
for level in range(tw.ms.nLevels):
if level==0:
tw.sample_gaussian(level,ms_Avees[level],ms_theta[level],mu=None)# zero mean
else:
tw.sample_from_the_ms_prior_coarse2fine_one_level(ms_Avees,ms_theta,
level_fine=level)
print('\nimg shape: {}\n'.format(img_original.shape))
# You don't have use these. You can use any 2d array
# that has two columns (regardless of the number of rows).
pts_src = tw.pts_src_dense
# Create buffers for the output
pts_fwd = CpuGpuArray.zeros_like(pts_src)
pts_inv = CpuGpuArray.zeros_like(pts_src)
for level in range(tw.ms.nLevels):
#######################################################################
# instead of the tw.sample_from_the_ms_prior() above,
# you may want to use one of the following.
# 1)
# tw.sample_gaussian(level,ms_Avees[level],ms_theta[level],mu=None)# zero mean
# 2)
# tw.sample_gaussian(level,ms_Avees[level],ms_theta[level],mu=some_user_specified_mu)
# The following should be used only for level>0 :
# 3)
# tw.sample_normal_in_one_level_using_the_coarser_as_mean(Avees_coarse=ms_Avees[level-1],
# Avees_fine=ms_Avees[level],
# theta_fine=ms_theta[level],
# level_fine=level)
#
#######################################################################
# You can also change the values this way:
# cpa_space = tw.ms.L_cpa_space[level]
# theta = cpa_space.get_zeros_theta()
# theta[:] = some values
# Avees = cpa_space.get_zeros_PA()
# cpa_space.theta2Avees(theta,Avees)
# cpa_space.update_pat(Avees)
# This step is important and must be done
# before are trying to "use" the new values of
# the (vectorized) A's.
tw.update_pat_from_Avees(ms_Avees[level],level)
if eval_v:
# Evaluating the velocity field.
# You don't have to do it in unless you want to visualize v.
# (when evaluting the treansformation, v will be internally
# evaluated anyway -- but its result won't be stored)
tw.calc_v(level=level)
# optional, if you want to time it
timer_gpu_T_fwd = GpuTimer()
# Simply calling
# tic = time.clock()
# and then
# tic = time.clock()
# won't work.
# In fact, most likely you will get that toc-tic is zero.
# You need to use the GpuTimer object. When you do that,
# one side effect is that suddenly the toc-tic from above will
# give you a more realistic result.
tic = time.clock()
timer_gpu_T_fwd.tic()
tw.calc_T_fwd(pts_src,pts_fwd,level=level)
timer_gpu_T_fwd.toc()
toc = time.clock()
print 'Time, in sec, for computing T_fwd:'
print timer_gpu_T_fwd.secs
print toc-tic # likely to be 0, unless you also used the GpuTimer.
# You can also time the inv of course. Results will be similar.
tw.calc_T_inv(pts_src,pts_inv,level=level)
if eval_cell_idx:
# cell_idx is computed here just for display.
cell_idx = CpuGpuArray.zeros(len(pts_src),dtype=np.int32)
tw.calc_cell_idx(pts_src,cell_idx,level,
permute_for_disp=permute_cell_idx_for_display)
# If may also want ro to time the remap.
# However, the remap is usually very fast (e.g, about 2 milisec).
# timer_gpu_remap_fwd = GpuTimer()
# tic = time.clock()
# timer_gpu_remap_fwd.tic()
# tw.remap_fwd(pts_inv=pts_inv,img=img_original,img_wrapped_fwd=img_wrapped_fwd)
tw.remap_fwd(pts_inv=pts_inv,img=img_original,img_wrapped_fwd=img_wrapped_fwd)
# timer_gpu_remap_fwd.toc()
# toc = time.clock()
# If the img type is np.float32, you may also use
# tw.remap_fwd_opencv instead of tw.remap_fw. The differences between
# the two methods are explained above
tw.remap_inv(pts_fwd=pts_fwd,img=img_original,img_wrapped_inv=img_wrapped_bwd)
# For display purposes, do gpu2cpu transfer
print ("For display purposes, do gpu2cpu transfer")
if eval_cell_idx:
cell_idx.gpu2cpu()
if eval_v:
tw.v_dense.gpu2cpu()
pts_fwd.gpu2cpu()
pts_inv.gpu2cpu()
img_wrapped_fwd.gpu2cpu()
img_wrapped_bwd.gpu2cpu()
figsize = (12,12)
plt.figure(figsize=figsize)
if eval_v:
plt.subplot(332)
tw.imshow_vx()
plt.title('vx')
plt.subplot(333)
tw.imshow_vy()
plt.title('vy')
if eval_cell_idx:
plt.subplot(331)
cell_idx_disp = cell_idx.cpu.reshape(img.shape[0],-1)
plt.imshow(cell_idx_disp)
plt.title('tess (type {})'.format(tess))
if show_downsampled_pts:
ds=20
pts_src_grid = pts_src.cpu.reshape(tw.nRows,-1,2)
pts_src_ds=pts_src_grid[::ds,::ds].reshape(-1,2)
pts_fwd_grid = pts_fwd.cpu.reshape(tw.nRows,-1,2)
pts_fwd_ds=pts_fwd_grid[::ds,::ds].reshape(-1,2)
pts_inv_grid = pts_inv.cpu.reshape(tw.nRows,-1,2)
pts_inv_ds=pts_inv_grid[::ds,::ds].reshape(-1,2)
use_lims=use_lims_when_plotting
# return tw
plt.subplot(334)
plt.plot(pts_src_ds[:,0],pts_src_ds[:,1],'r.')
plt.title('pts ds')
tw.config_plt()
plt.subplot(335)
plt.plot(pts_fwd_ds[:,0],pts_fwd_ds[:,1],'g.')
plt.title('fwd(pts)')
tw.config_plt(axis_on_or_off='on',use_lims=use_lims)
plt.subplot(336)
plt.plot(pts_inv_ds[:,0],pts_inv_ds[:,1],'b.')
plt.title('inv(pts)')
tw.config_plt(axis_on_or_off='on',use_lims=use_lims)
plt.subplot(337)
plt.imshow(img_original.cpu.astype(np.uint8))
plt.title('img')
# plt.axis('off')
plt.subplot(338)
plt.imshow(img_wrapped_fwd.cpu.astype(np.uint8))
# plt.axis('off')
plt.title('fwd(img)')
plt.subplot(339)
plt.imshow(img_wrapped_bwd.cpu.astype(np.uint8))
# plt.axis('off')
plt.title('inv(img)')
return tw
class Register(object):
def __init__(self,nRows=100,
nCols=100,
base = [2,2],
nLevels=4,
tess='tri',
zero_v_across_bdry=[False]*2,
# range_max_val = 1, # FOR NOW, KEEP IT.
# nPtsDense=10000,
scale_spatial=1.0 * 10,
scale_value=2.0,
sigma_signal=None,
wlp=1e-4,
ll_type=['gaussian','gaussian_on_distancetransform'][0],
only_local=False,
valid_outside=True):
ll_type = ll_type.lower()
if ll_type == 'gaussian':
self.SDLL=SDLL_gaussian
# elif ll_type =='gaussian_on_distancetransform':
# self.SDLL=SDLL_gaussian_on_distancetransform
# elif ll_type == 'vmf':
# self.SDLL=SDLL_VMF
else:
raise ValueError(ll_type)
# self.range_min_val = 0.0
# self.range_max_val = range_max_val
self.base = base
self.nLevels=nLevels
if sigma_signal is None:
raise ValueError("sigma_signal cannot be None")
self.sigma_signal = sigma_signal
self.wlp = wlp
self.tw = TransformWrapper(nRows=nRows,nCols=nCols,
nLevels=nLevels,
base=base,
tess=tess,
# nPtsDense=nPtsDense,
scale_spatial=scale_spatial,
scale_value=scale_value,
zero_v_across_bdry=zero_v_across_bdry,
only_local=only_local,
valid_outside=valid_outside
)
# 500,500,500,50000,5000
def set_dense(self,domain_start=-10,domain_end=10):
"""
Remarks:
1) The range of the domain has already been determined in self.tw
2) For now, the src is always the "uniform cdf"
"""
self.src_dense = self.tw.pts_src_dense
self.transformed_dense = self.tw.transformed_dense
def set_data(self,x,signal_src,signal_dst,isbinary):
"""
For now, assumes dst was evaluated on evenly-space points
// Is the comment above still current?
"""
self.isbinary=isbinary
nPts = len(x)
if x.ndim !=2 or x.shape[1]!=2:
raise ValueError(x.shape)
if signal_src.shape != signal_dst.shape:
raise ValueError(gnal_src.shape , signal_dst.shape)
if signal_src.ndim !=2:
# Want a single channel
raise ValueError(signal_src.shape)
# signal_src = signal_src.reshape(nPts,1).astype(np.float64)
# signal_dst = signal_dst.reshape(nPts,1).astype(np.float64)
signal_src = signal_src.astype(np.float64)
signal_dst = signal_dst.astype(np.float64)
# if nPts != signal_src.shape[0]:
# raise ValueError( nPts , signal_src.shape)
# if x.shape[0] != signal_dst.shape[0]:
# raise ValueError( nPts , signal_dst.shape)
if nPts != signal_src.size:
raise ValueError( nPts , signal_src.shape)
if x.shape[0] != signal_dst.size:
raise ValueError( nPts , signal_dst.shape)
if x.dtype != np.float64:
raise TypeError(x.dtype)
if signal_src.dtype != np.float64:
raise TypeError(signal_src.dtype)
if signal_dst.dtype != np.float64:
raise TypeError(signal_dst.dtype)
if signal_src.ndim == 1:
raise ValueError(signal_src.ndim)
if signal_dst.ndim == 1:
raise ValueError(signal_dst.ndim)
if not isinstance(signal_src,CpuGpuArray):
signal_src = CpuGpuArray(signal_src)
if not isinstance(signal_dst,CpuGpuArray):
signal_dst = CpuGpuArray(signal_dst)
self.signal = Bunch()
self.signal.src=signal_src
self.signal.dst=signal_dst
self.src = x
self.transformed = CpuGpuArray.zeros_like(self.src)
self.signal.transformed=CpuGpuArray.zeros_like(signal_src)
def set_run_lengths(self,run_lengths):
if len(run_lengths)!=self.nLevels:
raise ValueError(len(run_lengths),self.nLevels)
self.run_lengths = run_lengths
def fit(self,use_prior=False,proposal_scale=0.001,use_local=True,dispOn=True,
interp_type_for_ll=None,
interp_type_during_visualization=None,
scale_local_proposal=None):
nLevels=self.nLevels
tw = self.tw
sigma_signal = self.sigma_signal
self.use_local = use_local
inference_record = Bunch()
inference_record.tw_args = tw.args
inference_record.steps = []
inference_record.use_prior = use_prior
inference_record.proposal_scale= proposal_scale
inference_record.sigma_signal = sigma_signal
try:
run_lengths = self.run_lengths
except AttributeError:
self.set_run_lengths([500]*self.nLevels)
run_lengths = self.run_lengths
# raise Exception("self.set_run_lengths was not called yet")
wlp=self.wlp
for i in range(nLevels):
# for i in range(nLevels+5):
if i<nLevels:
level=i
if level == 0:
theta = tw.ms.L_cpa_space[level].get_zeros_theta()
else:
theta_fine = tw.ms.L_cpa_space[level].get_zeros_theta()
tw.ms.propogate_theta_coarse2fine(theta_coarse=theta,theta_fine=theta_fine)
theta = theta_fine
_sigma_signal = sigma_signal
else:
_sigma_signal *= 0.9
print '-'*10 ,'level',level, '-'*10
cpa_space = tw.ms.L_cpa_space[level]
print cpa_space
data = {'src':self.src,'transformed':self.transformed,
'signal':self.signal}
if use_prior:
lp_func = LP(ms=tw.ms,msp=tw.msp,level=level,SDLP=SDLP,
required={})
else:
lp_func = None
sampler = Metropolis(ll_func= LL(
ms=tw.ms,level=level,SDLL=self.SDLL,
data=data,
required={'sigma_signal':_sigma_signal,
'params_flow_int':tw.params_flow_int_coarse,
'interp_type_for_ll':interp_type_for_ll} ),
proposal=Proposal(ms=tw.ms,msp=tw.msp,level=level,
scale=proposal_scale,
use_local=use_local,
scale_local_proposal=scale_local_proposal),
# proposal=ProposalSphericalGaussian(ms=tw.ms,
# level=level,
# scale=0.1 / (1.2**level)
# scale=0.01,
# scale=0.1 / (1.1**level)
# scale=0.1
# ) ,
lp_func=lp_func,
wlp=wlp
)
sampler.set_theta(theta)
run_length = run_lengths[level]
sampler.run(run_length)
theta = sampler.theta_current # prepare for next iteration
inference_record.steps.append(sampler.get_record())
if dispOn:
if i >= nLevels-1 or 1:
plt.figure(i+1)
plt.clf()
self.disp(sampler=sampler,
interp_type_during_visualization=interp_type_during_visualization)
inference_record.theta = theta.copy()
steps = inference_record.steps
nAccepted = [ step.nAccepted for step in steps]
run_lengths = [ step.N for step in steps]
times = [ step.runs[0].time for step in steps]
total_time = sum(times)
print "run_lengths:",run_lengths
print "nAccepted:",nAccepted
print 'times:'
print_iterable(times)
print 'time total:',total_time
return theta.copy(),inference_record
def plot_src(self,*args,**kwargs):
x = self.x
plt.plot(x,self.src.cpu,*args,**kwargs)
# def plot_dst(self,*args,**kwargs):
# x = self.x
# plt.plot(x,self.dst.cpu,*args,**kwargs)
def __repr__(self):
s = 'Register:'
s += '\n'+ repr(self.tw.ms)
return s
@staticmethod
def plot_inference_summary(inference_record):
ll = []
lp = []
wlp_plus_ll=[]
for step in inference_record.steps:
ll += step.ll[1:] # start from 1 and not 0: to skip the initial guess
try:
lp += step.lp[1:]
wlp_plus_ll += list((step.wlp * np.asarray(step.lp[1:]) +
np.asarray(step.ll[1:])).tolist())
except AttributeError:
pass
plt.title('ll',fontsize=30)
plt.plot(ll,lw=2)
plt.plot(lp,lw=2)
plt.plot(wlp_plus_ll,lw=2)
counter = 0
for i,step in enumerate(inference_record.steps):
if i%2==1:
facecolor = ".2"
else:
facecolor = ".5"
plt.axvspan(counter, counter+step.nAccepted, facecolor=facecolor, alpha=0.2)
counter += step.nAccepted
def disp(self,sampler,interp_type_during_visualization):
level=sampler.level
theta=sampler.theta_current
tw=self.tw
# interval=self.interval
# interval_dense=self.interval_dense
markersize = 5
fontsize=30
cpa_space = tw.ms.L_cpa_space[level]
plt.subplot(231)
sampler.plot_ll()
plt.title('ll',fontsize=fontsize)
sampler.plot_wlp()
sampler.plot_wlp_plus_ll()
if sampler.lp_func:
plt.legend(['ll','wlp','ll+wlp'])
plt.subplot(232)
sampler.plot_ar()
plt.title('accept ratio',fontsize=fontsize)
# print theta
cpa_space.theta2As(theta=theta)
tw.update_pat_from_Avees(level=level)
tw.calc_v(level=level)
tw.v_dense.gpu2cpu()
src = self.src
# dst = self.dst
transformed = self.transformed
# src_dense=self.src_dense
# transformed_dense=self.transformed_dense
# tw.calc_T(src_dense, transformed_dense, mysign=1, level=level,
#
# transformed_dense.gpu2cpu()
tw.calc_T_inv(src, transformed, level=level,
int_quality=+1)
transformed.gpu2cpu()
if interp_type_during_visualization=='gpu_linear':
my_dtype = np.float64
else:
my_dtype = np.float32 # For opencv
img_src = self.signal.src.cpu.reshape(tw.nRows,tw.nCols)
img_src = CpuGpuArray(img_src.astype(my_dtype))
img_wrapped = CpuGpuArray.zeros_like(img_src)
img_dst = self.signal.dst.cpu.reshape(tw.nRows,tw.nCols)
img_dst = CpuGpuArray(img_dst)
if interp_type_during_visualization=='gpu_linear':
tw.remap_fwd(transformed,img_src,img_wrapped)
else:
tw.remap_fwd_opencv(transformed,img_src,img_wrapped,interp_type_during_visualization)
img_wrapped.gpu2cpu()
plt.subplot(233)
plt.imshow(img_src.cpu,interpolation="None")
plt.gray()
cpa_space.plot_cells('r')
tw.config_plt(axis_on_or_off='on')
plt.title(r'$I_{\mathrm{src}}$')
plt.subplot(234)
plt.imshow(img_wrapped.cpu,interpolation="None")
plt.gray()
# cpa_space.plot_cells('w')
tw.config_plt(axis_on_or_off='on')
plt.title(r'$I_{\mathrm{src}}\circ T^{\theta}$')
plt.subplot(235)
plt.imshow(img_dst.cpu,interpolation="None")
plt.gray()
plt.title(r'$I_{\mathrm{dst}}$')
# cpa_space.plot_cells('w')
tw.config_plt(axis_on_or_off='on')
plt.subplot(2,6,11)
self.tw.imshow_vx()
pylab.jet()
tw.config_plt(axis_on_or_off='on')
plt.title(r'$v_x$')
plt.subplot(2,6,12)
self.tw.imshow_vy()
pylab.jet()
tw.config_plt(axis_on_or_off='on')
plt.title(r'$v_y$')
pylab.ion()
name = 'LFW_5_to_6'
data = get_data(name)
src = CpuGpuArray(data.src)
dst = CpuGpuArray(data.dst)
transformed = CpuGpuArray.zeros_like(src)
fname_results = os.path.splitext(data.fname)[0]+'_result.pkl'
FilesDirs.raise_if_dir_does_not_exist(os.path.dirname(fname_results))
print 'Loading',fname_results
results=Pkl.load(fname_results)
theta_est = results.theta
tw = TransformWrapper(**results.tw_args)
tw.create_grid_lines(step=0.1,factor=0.5)
scale_quiver=1000 # The *smaller* this value is, the larger the plotted arrows will be.
level=-1 # pick the finest scale
cpa_space = tw.ms.L_cpa_space[level]
cpa_space.theta2Avees(theta_est)
cpa_space.update_pat()
tw.calc_T_fwd(src,transformed,level=level)
transformed.gpu2cpu()
tw.calc_v(level=level)
tw.v_dense.gpu2cpu()
plt.close('all')
disp(tw=tw,theta=theta_est,src=src,dst=dst,transformed=transformed,level=level,
if not inside_spyder():
pylab.ion()
name = 'LFW_5_to_6'
data = get_data(name)
src = CpuGpuArray(data.src)
dst = CpuGpuArray(data.dst)
transformed = CpuGpuArray.zeros_like(src)
fname_results = os.path.splitext(data.fname)[0] + '_result.pkl'
FilesDirs.raise_if_dir_does_not_exist(os.path.dirname(fname_results))
print 'Loading', fname_results
results = Pkl.load(fname_results)
theta_est = results.theta
tw = TransformWrapper(**results.tw_args)
tw.create_grid_lines(step=0.1, factor=0.5)
scale_quiver = 1000 # The *smaller* this value is, the larger the plotted arrows will be.
level = -1 # pick the finest scale
cpa_space = tw.ms.L_cpa_space[level]
cpa_space.theta2Avees(theta_est)
cpa_space.update_pat()
tw.calc_T_fwd(src, transformed, level=level)
transformed.gpu2cpu()
tw.calc_v(level=level)
tw.v_dense.gpu2cpu()
plt.close('all')
disp(tw=tw,
def example(img=None,
tess='I',
eval_cell_idx=True,
eval_v=True,
show_downsampled_pts=True,
valid_outside=True,
base=[1, 1],
scale_spatial=.1,
scale_value=100,
permute_cell_idx_for_display=True,
nLevels=3,
vol_preserve=False,
zero_v_across_bdry=[0, 0],
use_lims_when_plotting=True):
show_downsampled_pts = bool(show_downsampled_pts)
eval_cell_idx = bool(eval_cell_idx)
eval_v = bool(eval_cell_idx)
valid_outside = bool(valid_outside)
permute_cell_idx_for_display = bool(permute_cell_idx_for_display)
vol_preserve = bool(vol_preserve)
if img is None:
img = Img(get_std_test_img())
else:
img = Img(img)
img = img[:, :, ::-1] # bgr2rgb
tw = TransformWrapper(
nRows=img.shape[0],
nCols=img.shape[1],
nLevels=nLevels,
base=base,
scale_spatial=scale_spatial, # controls the prior's smoothness
scale_value=scale_value, # controls the prior's variance
tess=tess,
vol_preserve=vol_preserve,
zero_v_across_bdry=zero_v_across_bdry,
valid_outside=valid_outside)
print tw
# You probably want to do that: padding image border with zeros
border_width = 1
img[:border_width] = 0
img[-border_width:] = 0
img[:, :border_width] = 0
img[:, -border_width:] = 0
# The tw.calc_T_fwd (or tw.calc_T_inv) is always done in gpu.
# After using it to compute new pts,
# you may want to use remap (to warp an image accordingly).
# If you will use tw.remap_fwd (or tw.remap_inv), which is done in gpu,
# then the image type can be either float32 or float64.
# But if you plan to use tw.tw.remap_fwd_opencv (or tw.remap_inv_opencv),
# which is done in cpu (hence slightly lower) but supports better
# interpolation methods, then the image type must be np.float32.
# img_original = CpuGpuArray(img.copy().astype(np.float32))
img_original = CpuGpuArray(img.copy().astype(np.float64))
img_wrapped_fwd = CpuGpuArray.zeros_like(img_original)
img_wrapped_bwd = CpuGpuArray.zeros_like(img_original)
seed = 0
np.random.seed(seed)
ms_Avees = tw.get_zeros_PA_all_levels()
ms_theta = tw.get_zeros_theta_all_levels()
for level in range(tw.ms.nLevels):
if level == 0:
tw.sample_gaussian(level,
ms_Avees[level],
ms_theta[level],
mu=None) # zero mean
else:
tw.sample_from_the_ms_prior_coarse2fine_one_level(ms_Avees,
ms_theta,
level_fine=level)
print('\nimg shape: {}\n'.format(img_original.shape))
# You don't have use these. You can use any 2d array
# that has two columns (regardless of the number of rows).
pts_src = tw.pts_src_dense
# Create buffers for the output
pts_fwd = CpuGpuArray.zeros_like(pts_src)
pts_inv = CpuGpuArray.zeros_like(pts_src)
for level in range(tw.ms.nLevels):
#######################################################################
# instead of the tw.sample_from_the_ms_prior() above,
# you may want to use one of the following.
# 1)
# tw.sample_gaussian(level,ms_Avees[level],ms_theta[level],mu=None)# zero mean
# 2)
# tw.sample_gaussian(level,ms_Avees[level],ms_theta[level],mu=some_user_specified_mu)
# The following should be used only for level>0 :
# 3)
# tw.sample_normal_in_one_level_using_the_coarser_as_mean(Avees_coarse=ms_Avees[level-1],
# Avees_fine=ms_Avees[level],
# theta_fine=ms_theta[level],
# level_fine=level)
#
#######################################################################
# You can also change the values this way:
# cpa_space = tw.ms.L_cpa_space[level]
# theta = cpa_space.get_zeros_theta()
# theta[:] = some values
# Avees = cpa_space.get_zeros_PA()
# cpa_space.theta2Avees(theta,Avees)
# cpa_space.update_pat(Avees)
# This step is important and must be done
# before are trying to "use" the new values of
# the (vectorized) A's.
tw.update_pat_from_Avees(ms_Avees[level], level)
if eval_v:
# Evaluating the velocity field.
# You don't have to do it in unless you want to visualize v.
# (when evaluting the treansformation, v will be internally
# evaluated anyway -- but its result won't be stored)
tw.calc_v(level=level)
# optional, if you want to time it
timer_gpu_T_fwd = GpuTimer()
# Simply calling
# tic = time.clock()
# and then
# tic = time.clock()
# won't work.
# In fact, most likely you will get that toc-tic is zero.
# You need to use the GpuTimer object. When you do that,
# one side effect is that suddenly the toc-tic from above will
# give you a more realistic result.
tic = time.clock()
timer_gpu_T_fwd.tic()
tw.calc_T_fwd(pts_src, pts_fwd, level=level)
timer_gpu_T_fwd.toc()
toc = time.clock()
print 'Time, in sec, for computing T_fwd:'
print timer_gpu_T_fwd.secs
print toc - tic # likely to be 0, unless you also used the GpuTimer.
# You can also time the inv of course. Results will be similar.
tw.calc_T_inv(pts_src, pts_inv, level=level)
if eval_cell_idx:
# cell_idx is computed here just for display.
cell_idx = CpuGpuArray.zeros(len(pts_src), dtype=np.int32)
tw.calc_cell_idx(pts_src,
cell_idx,
level,
permute_for_disp=permute_cell_idx_for_display)
# If may also want ro to time the remap.
# However, the remap is usually very fast (e.g, about 2 milisec).
# timer_gpu_remap_fwd = GpuTimer()
# tic = time.clock()
# timer_gpu_remap_fwd.tic()
# tw.remap_fwd(pts_inv=pts_inv,img=img_original,img_wrapped_fwd=img_wrapped_fwd)
tw.remap_fwd(pts_inv=pts_inv,
img=img_original,
img_wrapped_fwd=img_wrapped_fwd)
# timer_gpu_remap_fwd.toc()
# toc = time.clock()
# If the img type is np.float32, you may also use
# tw.remap_fwd_opencv instead of tw.remap_fw. The differences between
# the two methods are explained above
tw.remap_inv(pts_fwd=pts_fwd,
img=img_original,
img_wrapped_inv=img_wrapped_bwd)
# For display purposes, do gpu2cpu transfer
print("For display purposes, do gpu2cpu transfer")
if eval_cell_idx:
cell_idx.gpu2cpu()
if eval_v:
tw.v_dense.gpu2cpu()
pts_fwd.gpu2cpu()
pts_inv.gpu2cpu()
img_wrapped_fwd.gpu2cpu()
img_wrapped_bwd.gpu2cpu()
figsize = (12, 12)
plt.figure(figsize=figsize)
if eval_v:
plt.subplot(332)
tw.imshow_vx()
plt.title('vx')
plt.subplot(333)
tw.imshow_vy()
plt.title('vy')
if eval_cell_idx:
plt.subplot(331)
cell_idx_disp = cell_idx.cpu.reshape(img.shape[0], -1)
plt.imshow(cell_idx_disp)
plt.title('tess (type {})'.format(tess))
if show_downsampled_pts:
ds = 20
pts_src_grid = pts_src.cpu.reshape(tw.nRows, -1, 2)
pts_src_ds = pts_src_grid[::ds, ::ds].reshape(-1, 2)
pts_fwd_grid = pts_fwd.cpu.reshape(tw.nRows, -1, 2)
pts_fwd_ds = pts_fwd_grid[::ds, ::ds].reshape(-1, 2)
pts_inv_grid = pts_inv.cpu.reshape(tw.nRows, -1, 2)
pts_inv_ds = pts_inv_grid[::ds, ::ds].reshape(-1, 2)
use_lims = use_lims_when_plotting
# return tw
plt.subplot(334)
plt.plot(pts_src_ds[:, 0], pts_src_ds[:, 1], 'r.')
plt.title('pts ds')
tw.config_plt()
plt.subplot(335)
plt.plot(pts_fwd_ds[:, 0], pts_fwd_ds[:, 1], 'g.')
plt.title('fwd(pts)')
tw.config_plt(axis_on_or_off='on', use_lims=use_lims)
plt.subplot(336)
plt.plot(pts_inv_ds[:, 0], pts_inv_ds[:, 1], 'b.')
plt.title('inv(pts)')
tw.config_plt(axis_on_or_off='on', use_lims=use_lims)
plt.subplot(337)
plt.imshow(img_original.cpu.astype(np.uint8))
plt.title('img')
# plt.axis('off')
plt.subplot(338)
plt.imshow(img_wrapped_fwd.cpu.astype(np.uint8))
# plt.axis('off')
plt.title('fwd(img)')
plt.subplot(339)
plt.imshow(img_wrapped_bwd.cpu.astype(np.uint8))
# plt.axis('off')
plt.title('inv(img)')
return tw