val_output_tmp = model_512.predict([
val_90d[:, ::img_scale_inv, ::img_scale_inv],
val_0d[:, ::img_scale_inv, ::img_scale_inv],
val_M45d[:, ::img_scale_inv, ::img_scale_inv]
],
batch_size=1)
if corner_code == "NE":
val_output_tmp = np.rot90(val_output_tmp, axes=(2, 1))
elif corner_code == "SW":
val_output_tmp = np.rot90(val_output_tmp, axes=(1, 2))
elif corner_code == "SE":
val_output_tmp = np.rot90(val_output_tmp, k=2, axes=(2, 1))
runtime = time.clock() - start
plt.imshow(val_output_tmp[0, :, :, 0])
plt.imsave(
output_dir + sample + "_" + str(corner_coords[0]) + "_" +
str(corner_coords[1]) + '.png', val_output_tmp[0, :, :, 0])
plt.show()
print("runtime: %.5f(s)" % runtime)
# save .pfm file
pfm_path = output_dir + "pfms/" + sample + "_" + str(
corner_coords[0]) + "_" + str(corner_coords[1]) + '.pfm'
write_pfm(val_output_tmp[0, :, :, 0], pfm_path)
print('pfm file saved in ' + pfm_path)
dummy = model_512.predict([dum, dum, dum, dum], batch_size=1)
""" Depth Estimation """
for image_path in dir_LFimages:
(val_90d, val_0d, val_45d,
val_M45d) = make_multiinput(image_path, image_h, image_w,
Setting02_AngualrViews)
# print(val_90d.shape)
# exit()
start = time.time()
# predict
val_output_tmp = model_512.predict([
val_90d[:, ::img_scale_inv, ::img_scale_inv],
val_0d[:, ::img_scale_inv, ::img_scale_inv],
val_45d[:, ::img_scale_inv, ::img_scale_inv],
val_M45d[:, ::img_scale_inv, ::img_scale_inv]
],
batch_size=1)
runtime = time.time() - start
plt.imshow(val_output_tmp[0, :, :, 0])
print("runtime: %.5f(s)" % runtime)
# save .pfm file
write_pfm(val_output_tmp[0, :, :, 0],
dir_output + '/%s.pfm' % (image_path.split('/')[-1]))
print('pfm file saved in %s/%s.pfm' %
(dir_output, image_path.split('/')[-1]))
show_warped(Warped[:, :, 0])
Warpeds[corner_coord] = Warped
### read the center:
center_pfms_dir = center_res_dir + 'pfms/'
print("doing.." + str(center_view))
center_D = read_pfm(glob(center_pfms_dir + '*.pfm')[0])
Warpeds[center_view] = np.expand_dims(center_D, -1)
#### fuse:
Warped_ims = np.concatenate(list(Warpeds.values()), -1)
#fused_Warped = np.median(Warped_ims*(Warped_ims!=inv_pix_val),-1)
fused_Warped = np.zeros_like(center_D)
for ir in range(D.shape[0]):
for ic in range(D.shape[1]):
vals = Warped_ims[ir, ic, :]
valid_vals = vals[vals != inv_pix_val]
fused_Warped[ir, ic] = np.median(valid_vals)
## show end result
plt_imshow(center_D)
plt_imshow(fused_Warped)
if save_output:
write_pfm(
fused_Warped, output_dir + sample + '_' + str(center_view[0]) + '_' +
str(center_view[1]) + '.pfm')
