def get_gaussian_weight_patch(gauss_shape=(19, 19), gauss_sigma_frac=.3,
gauss_norm_01=True):
r"""
2d gaussian image useful for plotting
Returns:
ndarray: patch
CommandLine:
python -m vtool.coverage_kpts --test-get_gaussian_weight_patch
Example:
>>> # ENABLE_DOCTEST
>>> from vtool.coverage_kpts import * # NOQA
>>> # build test data
>>> # execute function
>>> patch = get_gaussian_weight_patch()
>>> # verify results
>>> result = str(patch)
>>> print(result)
"""
# Perdoch uses roughly .95 of the radius
radius = gauss_shape[0] / 2.0
sigma = gauss_sigma_frac * radius
# Similar to SIFT's computeCircularGaussMask in helpers.cpp
# uses smmWindowSize=19 in hesaff for patch size. and 1.6 for sigma
# Create gaussian image to warp
patch = ptool.gaussian_patch(shape=gauss_shape, sigma=sigma)
if gauss_norm_01:
np.divide(patch, patch.max(), out=patch)
return patch
def get_gaussian_weight_patch(
gauss_shape=(19, 19), gauss_sigma_frac=.3, gauss_norm_01=True):
r"""
2d gaussian image useful for plotting
Returns:
ndarray: patch
CommandLine:
python -m vtool.coverage_kpts --test-get_gaussian_weight_patch
Example:
>>> # ENABLE_DOCTEST
>>> from vtool.coverage_kpts import * # NOQA
>>> # build test data
>>> # execute function
>>> patch = get_gaussian_weight_patch()
>>> # verify results
>>> result = str(patch)
>>> print(result)
"""
# Perdoch uses roughly .95 of the radius
radius = gauss_shape[0] / 2.0
sigma = gauss_sigma_frac * radius
# Similar to SIFT's computeCircularGaussMask in helpers.cpp
# uses smmWindowSize=19 in hesaff for patch size. and 1.6 for sigma
# Create gaussian image to warp
patch = ptool.gaussian_patch(shape=gauss_shape, sigma=sigma)
if gauss_norm_01:
np.divide(patch, patch.max(), out=patch)
return patch
def TEST_figure1(wpatch, gradx, grady, gmag, gori, hist, centers):
print('[rotinvar] 4) Draw histogram with interpolation annotations')
fnum = 1
gorimag = plottool.color_orimag(gori, gmag, True)
nRow, nCol = (2, 7)
df2.figure(fnum=1, pnum=(nRow, 1, nRow), doclf=True, docla=True)
plottool.draw_hist_subbin_maxima(hist, centers)
df2.set_xlabel('grad orientation (radians)')
df2.set_ylabel('grad magnitude')
df2.set_title('dominant orientations')
print('[rotinvar] 5) Show patch, gradients, magintude, and orientation')
df2.imshow(wpatch, pnum=(nRow, nCol, 1), fnum=fnum, title='patch')
df2.draw_vector_field(gradx, grady, pnum=(nRow, nCol, 2), fnum=fnum, title='gori (vec)')
df2.imshow(gorimag, pnum=(nRow, nCol, 3), fnum=fnum, title='gori (col)')
df2.imshow(np.abs(gradx), pnum=(nRow, nCol, 4), fnum=fnum, title='gradx')
df2.imshow(np.abs(grady), pnum=(nRow, nCol, 5), fnum=fnum, title='grady')
df2.imshow(gmag, pnum=(nRow, nCol, 6), fnum=fnum, title='gmag')
gpatch = ptool.gaussian_patch(shape=gori.shape)
df2.imshow(gpatch * 255, pnum=(nRow, nCol, 7), fnum=fnum, title='gauss weights', cmap_='hot')
#gpatch3 = np.dstack((gpatch, gpatch, gpatch))
#df2.draw_vector_field(gradx * gpatch, grady * gpatch, pnum=(nRow, nCol, 8), fnum=fnum, title='gori (vec)')
#df2.imshow(gorimag * gpatch3, pnum=(nRow, nCol, 9), fnum=fnum, title='gori (col)')
#df2.imshow(gradx * gpatch, pnum=(nRow, nCol, 10), fnum=fnum, title='gradx')
#df2.imshow(grady * gpatch, pnum=(nRow, nCol, 11), fnum=fnum, title='grady')
#df2.imshow(gmag * gpatch, pnum=(nRow, nCol, 12), fnum=fnum, title='gmag')
return locals()
def get_annot_match_covimg(ibs, aid, sel_fx, mx2_score, **kwargs):
chip = ibs.get_annot_chips(aid)
kpts = ibs.get_annot_kpts(aid)
mx2_kp = kpts[sel_fx]
srcimg = ptool.gaussian_patch()
# 2 and 3 are scale modes
if kwargs.get('method', 0) in [2, 3]:
# Bigger keypoints should get smaller weights
mx2_scale = np.sqrt([a * d for (x, y, a, c, d) in mx2_kp])
mx2_score = mx2_score / mx2_scale
dstimg = warp_srcimg_to_kpts(mx2_kp,
srcimg,
chip.shape[0:2],
fx2_score=mx2_score,
**kwargs)
return dstimg
def get_coverage_map(kpts, chip_shape, **kwargs):
# Create gaussian image to warp
np.tau = 2 * np.pi
srcimg = ptool.gaussian_patch()
dstimg = warp_srcimg_to_kpts(kpts, srcimg, chip_shape, **kwargs)
return dstimg
