pr = get_prior(prior, target_pixels)
backward_ms = ms.copy()
for idx in range(len(target_pixels)):
p = target_pixels[idx]
backward_ms[idx, p[0] + 1:, :] = 1
print(np.sum(1 - backward_ms[0]))
for idx in range(len(target_pixels)):
p = target_pixels[idx]
backward_ms[idx, p[0], p[1]] = 1
backward_ms = np.rot90(backward_ms, 2, (1, 2))
# Forward model prediction
#feed_dict = fm.make_feed_dict(d, mask_values=ams, rot=False)
feed_dict = fm.make_feed_dict(d,
mask_values=np.rot90(
backward_ms, 2, (1, 2)),
rot=False)
o1 = sess.run(fm.outputs, feed_dict)
o1 = np.concatenate(o1, axis=0)
# coeffs, means, stds = transform_params(o1, fm.args.nr_logistic_mix)
o1 = get_params(o1, target_pixels)
# c = coeffs[:, :, :, :]
# s = stds[:, :, :, 0, :] * 127.5
# r = np.sum(c * s, axis=-1)
# r = np.mean(r, axis=0)
# r *= (1-ms[0])
# print(r[28:36,28:36])
# print("----------------")
# Backward model prediction
#feed_dict = bm.make_feed_dict(d, mask_values=backward_ms, rot=True)
while True:
count += 1
#print(count)
target_pixels = next_pixel(ms)
#print(target_pixels[0])
if target_pixels[0][0] is None:
break
pr = get_prior(prior, target_pixels)
backward_ms = ms.copy()
for idx in range(len(target_pixels)):
p = target_pixels[idx]
backward_ms[idx, p[0], p[1]] = 1
backward_ms = np.rot90(ms, 2, (1, 2))
# Forward model prediction
feed_dict = fm.make_feed_dict(d, mask_values=ams, rot=False)
o1 = sess.run(fm.outputs, feed_dict)
o1 = np.concatenate(o1, axis=0)
o1 = get_params(o1, target_pixels)
# Backward model prediction
feed_dict = bm.make_feed_dict(d,
mask_values=backward_ms,
rot=True)
o2 = sess.run(bm.outputs, feed_dict)
o2 = np.concatenate(o2, axis=0)
o2 = np.rot90(o2, 2, (1, 2))
o2 = get_params(o2, target_pixels)
# Sample red channel
pars1 = params_to_dis(o1, fm.args.nr_logistic_mix)
