def make_fine(idrive, filename=None):
if filename is None:
filename = "drive%02d_fine.npy" % idrive
video = load_stereo_video(idrive)
# video = video[:3]
n_disp = 128
fine_params = {
'data_weight': 0.01289,
'disc_max': 411.9,
'data_max': 2382540.,
'ksize': 15,
'iters': 10
} # 2.023
# disps = []
disps = np.zeros((len(video), ) + video[0][0].shape, dtype='uint8')
for i, frame in enumerate(video):
tic('BP frame %d' % i)
disp = fine_bp(frame, values=n_disp, **fine_params)
toc()
# print disp.dtype
# disps.append(disp)
disps[i] = disp
# disps = np.array(disps)
np.save(filename, disps)
def estimate_coarse_error(down_factor, iters, video, n_disp, drive, n_frames=20, **params):
# TODO: use fine BP as ground truth?
coarse_error = []
for i in range(n_frames):
print(i)
frame = video[i]
tic('coarse')
coarse_disp = coarse_bp(frame, down_factor=down_factor, iters=iters, **params)
toc()
xyd = load_disparity_points(drive, i)
coarse_error.append(error_on_points(xyd, coarse_disp, n_disp, kind='close'))
return np.mean(coarse_error)
def test_tic_toc():
rng = np.random.RandomState(0)
n = int(1e6)
tic()
rng.normal(size=n)
t0 = toc()
tic()
rng.normal(size=5 * n)
t1 = toc()
assert 0.001 <= t0 <= 10
assert 3 * t0 <= t1 <= 30
def test_shift_images():
# n = 5
# images = np.zeros((n, 9))
# shape = (3, 3)
# images[np.arange(n), 4] = 1
# images2 = shift_images(images, shape)
# for image in images2:
# print image.reshape(shape)
# --- timing test
from hunse_tools.timing import tic, toc
[train_images, _], _, _ = mnist()
tic()
images2 = shift_images(train_images, (28, 28))
toc()
def test_shift_images():
# n = 5
# images = np.zeros((n, 9))
# shape = (3, 3)
# images[np.arange(n), 4] = 1
# images2 = shift_images(images, shape)
# for image in images2:
# print image.reshape(shape)
# --- timing test
from hunse_tools.timing import tic, toc
[train_images, _], _, _ = mnist()
tic()
images2 = shift_images(train_images, (28, 28))
toc()
a = nengo.Ensemble(n_hid,
n_vis,
eval_points=images,
encoders=encoders,
intercepts=[-0.5] * n_hid,
max_rates=[200] * n_hid)
o = nengo.Node(size_in=n_vis)
op = nengo.Probe(o, synapse=0.03)
ca = nengo.Connection(u, a)
co = nengo.Connection(a, o)
from hunse_tools.timing import tic, toc
tic()
sim = nengo.Simulator(model)
toc()
tic()
sim.run(1.0)
toc()
x = sim.data[up].reshape(-1, 28, 28)
y = sim.data[op].reshape(-1, 28, 28)
plt.figure(1)
plt.clf()
plt.subplot(121)
plt.imshow(x.mean(0), cmap='gray')
plt.subplot(122)
plt.imshow(y.mean(0), cmap='gray')
frame = load_stereo_frame(idrive, iframe)
points = load_disparity_points(idrive, iframe)
full_values = 128
frame_down_factor = 1
values = full_values / 2**frame_down_factor
iters = 3
params = {
'data_weight': 0.16145115747533928, 'disc_max': 294.1504935618425,
'data_max': 32.024780646200725, 'ksize': 3}
frame = (downsample(frame[0], frame_down_factor),
downsample(frame[1], frame_down_factor))
tic()
coarse_disp = coarse_bp(frame, values=values, down_factor=1, iters=iters, **params)
coarse_disp *= 2**frame_down_factor
toc()
tic()
fine_disp = coarse_bp(frame, values=values, down_factor=0, iters=iters, **params)
fine_disp *= 2**frame_down_factor
toc()
# fovea_corners = (60, 360)
# fovea_shapes = (80, 80)
fovea_corners = [(60, 160), (60, 360)]
fovea_shapes = [(80, 80), (80, 80)]
fovea_corners = np.array(fovea_corners, ndmin=2)
coarse_shape = low0.shape
fine_shape = np.asarray(high0.shape)
filt = BryanFilter(coarse_shape, fine_shape, fovea_shape)
# filt.fovea_margin = fovea_margin
# filt_nofovea = BryanFilter(coarse_shape, fovea_shape)
fig = plt.figure(1)
fig.clf()
ax_disp = plt.gca()
plot_disp = ax_disp.imshow(high1, vmin=0, vmax=n_disp)
for iframe, [frame, pos] in enumerate(zip(video, positions)):
tic('Coarse BP')
coarse_disp = coarse_bp(frame, down_factor=coarse_down_factor, iters=coarse_iters, **coarse_params)
toc()
ratio = np.round(float(frame.shape[1]) / float(coarse_disp.shape[0]))
ij0 = np.round(np.asarray(fovea_ij) / ratio)
coarse_shape = np.round(np.asarray(fovea_shape) / ratio)
ij1 = ij0 + coarse_shape
coarse_subwindow = coarse_disp[ij0[0]:ij1[0], ij0[1]:ij1[1]]
# TODO: scaling only the edges may be faster
seed = cv2.resize(coarse_subwindow, fovea_shape[::-1])
band_size = 10
seed[band_size:-band_size, band_size:-band_size] = 0
tic('Fine BP')
# plot_disp2 = ax_disp2.imshow(high1, vmin=0, vmax=n_disp)
# plot_truth = ax_truth.imshow(low0, vmin=0, vmax=n_disp)
# ax_cost.imshow(filt.cost) # for tight layout
#
# ax_frame.set_title('Left camera frame (red: current fovea, green: next fovea)')
# ax_coarse.set_title('Coarse')
# ax_fovea.set_title('Fovea')
# ax_disp.set_title('Estimated disparity (fovea)')
# ax_disp2.set_title('Estimated disparity (no fovea)')
# ax_cost.set_title('Estimated cost')
# ax_truth.set_title('Truth')
# fig.tight_layout()
for iframe, [frame, pos] in enumerate(zip(video, positions)):
tic('Coarse BP')
coarse_disp = coarse_bp(frame, **coarse_params)
toc()
ratio = np.round(float(frame.shape[1]) / float(coarse_disp.shape[0]))
ij0 = np.round(np.asarray(fovea_ij) / ratio)
coarse_shape = np.round(np.asarray(fovea_shape) / ratio)
ij1 = ij0 + coarse_shape
coarse_subwindow = coarse_disp[ij0[0]:ij1[0], ij0[1]:ij1[1]]
# TODO: scaling only the edges may be faster
seed = cv2.resize(coarse_subwindow, fovea_shape[::-1])
band_size = 20
seed[band_size:-band_size, band_size:-band_size] = 0
tic('Fine BP')