def create_augumentation_mesh(self, dim, newdim, generate_type):
"""
return a list of offset
[(dr1,dc1),(dr2,dc2), ...,(drn,dcn)]
(dri,dci) is the lower points for i-th window( in row, col format)
"""
if generate_type in set(['rt', 'neg_sample']):
mesh = iu.cartesian_product2(np.asarray( [range(dim[0] - newdim[0] + 1), range(dim[1] - newdim[1] + 1)]))
elif generate_type == 'ct':
dx = (dim[0] - newdim[0])/2
dy = (dim[1] - newdim[1])/2
mesh =[(dx,dy)]
return mesh
def create_augumentation_mesh(self, dim, newdim, generate_type):
"""
return a list of offset
[(dr1,dc1),(dr2,dc2), ...,(drn,dcn)]
(dri,dci) is the lower points for i-th window( in row, col format)
"""
if generate_type in set(['rt', 'neg_sample']):
mesh = iu.cartesian_product2(
np.asarray([
range(dim[0] - newdim[0] + 1),
range(dim[1] - newdim[1] + 1)
]))
elif generate_type == 'ct':
dx = (dim[0] - newdim[0]) / 2
dy = (dim[1] - newdim[1]) / 2
mesh = [(dx, dy)]
return mesh
def generate_rt_data(self, matpath):
"""
in each mat
mat['X'] is image data
mat['Y'] is npart x ndata array
"""
mat = sio.loadmat(matpath)
dim = mat['dim'][0]
ndata = (mat['X'].shape)[1]
sample_num = self.savedata_info['sample_num']
totaldata = sample_num * ndata * 2
newdim = self.savedata_info['newdim']
nparts = self.meta['nparts']
filter_size = self.savedata_info['indmap_para']['filter_size']
stride = self.savedata_info['indmap_para']['stride']
rate = self.savedata_info['indmap_para']['rate']
mdim = self.get_indmapdim(newdim, filter_size, stride)
if newdim[0] > dim[0] or newdim[1] > dim[1]:
raise HMLPEError('Invalid new size ')
if (dim[0] - newdim[0] + 1) * (dim[1] - newdim[1] + 1) < sample_num:
raise HMLPEError(' Invalid sample_num')
res = {}
per_size = min(totaldata, self.savedata_info['max_batch_size'])
mesh = iu.cartesian_product2(np.asarray( [range(dim[0] - newdim[0] + 1), \
range(dim[1] - newdim[1] + 1)]))
allX = mat['X'].reshape( (dim[0], dim[1],dim[2], ndata), order='F')
allY = mat['Y'].reshape( (2,-1, ndata), order='F')
newlen = iu.prod( newdim )
res['data'] = np.ndarray([newdim[0], newdim[1], newdim[2], per_size], \
dtype = np.uint8)
res['labels'] = np.ndarray( [allY.shape[1], per_size], dtype=np.int)
res['joints8'] = np.ndarray( [8, 2, per_size], dtype=np.float)
res['oribbox'] = np.ndarray( [4, per_size], dtype = np.float)
res['indmap'] = np.zeros((nparts, mdim[0], mdim[1], per_size), np.bool)
res['filenames'] =[str() for x in range(per_size)]
tmpres = dict()
cur_id = self.batch_id
pre_nc = 0
nc = 0
for it in range(ndata):
curX = allX[...,it]
curY = allY[...,it].transpose()
curfilename = str(mat['imagepathlist'][0,it][0])
l = (np.random.permutation(range(len(mesh))))[:sample_num]
for p in l:
r,c = mesh[p]
tmpX = curX
tmpX = np.roll(tmpX, shift=-int(r), axis = 0)
tmpX = np.roll(tmpX, shift=-int(c), axis = 1)
tmpX = tmpX[:newdim[0], :newdim[1],:]
res['data'][...,nc - pre_nc] = tmpX
tmpY = curY - 1 + np.asarray([-c,-r])
res['joints8'][..., nc - pre_nc] = tmpY;
res['filenames'][nc - pre_nc] = curfilename
res['oribbox'][...,nc-pre_nc] = res['oribbox'][...,it]
res['indmap'][...,nc-pre_nc] = self.create_indicatormap(tmpY, part_idx, mdim, rate, filter_size, stride)
res['jointmasks'] = self.makejointmask(newdim, tmpY)
nc = nc + 1
#flip image
tmpX = tmpX[:,::-1,:]
res['data'][...,nc - pre_nc] = tmpX
res['filenames'][nc - pre_nc] = curfilename
tmmY = flip_pose8_joint(newdim, tmpY)
res['joints8'][...,nc -pre_nc] = tmpY
res['oribbox'][...,nc-pre_nc] = res['oribbox'][...,it]
res['indmap'][...,nc-pre_nc] = self.create_indicatormap(tmpY, part_idx, mdim, rate, filter_size, stride)
res['jointmasks'] = self.makejointmask(newdim, tmpY)
