Skip to content

Commit af73737

Browse files
author
prabhatkc
committed
updating due to the addition of conventional artifacts
1 parent 9a499c0 commit af73737

23 files changed

+387
-476
lines changed

‎paper_plots/banded_plots.py‎

Lines changed: 260 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,260 @@
1+
##############################################################
2+
# @author: pkc
3+
#
4+
# banded_plots.py
5+
# ............
6+
# includes codes to output different banded plots shown in the
7+
# sfrc paper
8+
# comment the line mpi.use('Agg') in the file src/plot_func.py
9+
# to view plot as display
10+
11+
import sys
12+
sys.path.append('../')
13+
import numpy as np
14+
import numpy.fft as fft
15+
from src import frc_utils
16+
from src import io_func
17+
from src import frc_utils
18+
from src import plot_func as pf
19+
import os
20+
from src import utils
21+
22+
def get3_comp_of_img(img, band1_ep, band2_ep, band3_ep):
23+
h, w = img.shape
24+
img_p1 = np.zeros([h,w]).astype('complex')
25+
img_p2 = np.zeros([h,w]).astype('complex')
26+
img_p3 = np.zeros([h,w]).astype('complex')
27+
img_fft = fft.fftshift(fft.fft2(img))
28+
29+
indices=frc_utils.ring_indices(img_fft)
30+
for i in range(0, band1_ep):
31+
img_p1[indices[i]]=img_fft[indices[i]]
32+
for i in range(band1_ep, band2_ep):
33+
img_p2[indices[i]]=img_fft[indices[i]]
34+
for i in range(band2_ep, band3_ep):
35+
img_p3[indices[i]]=img_fft[indices[i]]
36+
37+
img_p1 = np.real(fft.ifft2(fft.ifftshift(img_p1)))
38+
img_p2 = np.real(fft.ifft2(fft.ifftshift(img_p2)))
39+
img_p3 = np.real(fft.ifft2(fft.ifftshift(img_p3)))
40+
img_stack = np.stack((img.reshape(h,w), img_p1.reshape(h, w), img_p2.reshape(h, w), img_p3.reshape(h, w)), axis=0)
41+
return(img_p1, img_p2, img_p3, img_stack)
42+
43+
def get3_fft_bands_of_img(img, band1_ep, band2_ep, band3_ep):
44+
h, w = img.shape
45+
img_p1 = np.zeros([h,w])
46+
img_p2 = np.zeros([h,w])
47+
img_p3 = np.zeros([h,w])
48+
img_fft = np.log(np.abs(fft.fftshift(fft.fft2(img))))
49+
50+
indices=frc_utils.ring_indices(img_fft)
51+
for i in range(0, band1_ep):
52+
img_p1[indices[i]]=img_fft[indices[i]]
53+
for i in range(band1_ep, band2_ep):
54+
img_p2[indices[i]]=img_fft[indices[i]]
55+
for i in range(band2_ep, band3_ep):
56+
img_p3[indices[i]]=img_fft[indices[i]]
57+
img_stack = np.stack((img_fft.reshape(h,w), img_p1.reshape(h, w), img_p2.reshape(h, w), img_p3.reshape(h, w)), axis=0)
58+
return(img_p1, img_p2, img_p3, img_stack)
59+
60+
def get5_comp_of_img(img, band1_ep, band2_ep, band3_ep, band4_ep, band5_ep):
61+
h, w = img.shape
62+
img_p1 = np.zeros([h,w]).astype('complex')
63+
img_p2 = np.zeros([h,w]).astype('complex')
64+
img_p3 = np.zeros([h,w]).astype('complex')
65+
img_p4 = np.zeros([h,w]).astype('complex')
66+
img_p5 = np.zeros([h,w]).astype('complex')
67+
img_fft = fft.fftshift(fft.fft2(img))
68+
69+
indices=frc_utils.ring_indices(img_fft)
70+
for i in range(0, band1_ep):
71+
img_p1[indices[i]]=img_fft[indices[i]]
72+
for i in range(band1_ep, band2_ep):
73+
img_p2[indices[i]]=img_fft[indices[i]]
74+
for i in range(band2_ep, band3_ep):
75+
img_p3[indices[i]]=img_fft[indices[i]]
76+
for i in range(band3_ep, band4_ep):
77+
img_p4[indices[i]]=img_fft[indices[i]]
78+
for i in range(band4_ep, band5_ep):
79+
img_p5[indices[i]]=img_fft[indices[i]]
80+
81+
img_p1 = np.real(fft.ifft2(fft.ifftshift(img_p1)))
82+
img_p2 = np.real(fft.ifft2(fft.ifftshift(img_p2)))
83+
img_p3 = np.real(fft.ifft2(fft.ifftshift(img_p3)))
84+
img_p4 = np.real(fft.ifft2(fft.ifftshift(img_p4)))
85+
img_p5 = np.real(fft.ifft2(fft.ifftshift(img_p5)))
86+
img_stack = np.stack((img.reshape(h,w), img_p1.reshape(h, w), img_p2.reshape(h, w), img_p3.reshape(h, w), img_p4.reshape(h, w), img_p5.reshape(h, w)), axis=0)
87+
return(img_p1, img_p2, img_p3, img_p4, img_p5, img_stack)
88+
89+
def get5_fft_bands_of_img(img, band1_ep, band2_ep, band3_ep, band4_ep, band5_ep):
90+
h, w = img.shape
91+
img_p1 = np.zeros([h,w])
92+
img_p2 = np.zeros([h,w])
93+
img_p3 = np.zeros([h,w])
94+
img_p4 = np.zeros([h,w])
95+
img_p5 = np.zeros([h,w])
96+
img_fft = np.log(np.abs(fft.fftshift(fft.fft2(img))))
97+
98+
indices=frc_utils.ring_indices(img_fft)
99+
for i in range(0, band1_ep):
100+
img_p1[indices[i]]=img_fft[indices[i]]
101+
for i in range(band1_ep, band2_ep):
102+
img_p2[indices[i]]=img_fft[indices[i]]
103+
for i in range(band2_ep, band3_ep):
104+
img_p3[indices[i]]=img_fft[indices[i]]
105+
for i in range(band3_ep, band4_ep):
106+
img_p4[indices[i]]=img_fft[indices[i]]
107+
for i in range(band4_ep, band5_ep):
108+
img_p5[indices[i]]=img_fft[indices[i]]
109+
img_stack = np.stack((img_fft.reshape(h,w), img_p1.reshape(h, w), img_p2.reshape(h, w), img_p3.reshape(h, w), img_p4.reshape(h, w), img_p5.reshape(h, w) ), axis=0)
110+
return(img_p1, img_p2, img_p3, img_p4, img_p5, img_stack)
111+
# --------------------------------------
112+
# Display and saving figure option
113+
# -----------------------------------------
114+
plot_fig = True # Display plots
115+
save_fig = False # save plots
116+
crop_fig = True # banded plots related cropped image in the main paper vs full image in the supp paper
117+
Ncomp = 5 # no. of binned fft components. Other option is 3
118+
119+
if crop_fig:
120+
# -----------------------------------------------------------------
121+
# this option corresponds to banded plots of the cropped ROIs
122+
# between FBP and SRGAN shown in the main paper.
123+
# cnn is pointed to SRGAN-based outputs
124+
# -----------------------------------------------------------------
125+
gt_path = './plot2/crop_img_uint8_L_50_W_400/gt_000069.png'
126+
cnn_path = './plot2/crop_img_uint8_L_50_W_400/srgan_000069.png' #SRGAN outputs
127+
out_path = './plot2/crop_fig/Ncomp_'+ str(Ncomp)+ '/'
128+
else:
129+
# -----------------------------------------------------------------
130+
# this option corresponds to the banded plots of full images
131+
# related to conventional artifacts showns in supplemental document.
132+
# cnn is pointed to missing-wedge-based outputs.
133+
# -----------------------------------------------------------------
134+
#gt_path = '../nfk_artifacts/data/missing-wedge/gt_irt/mk_L291_tv_000163_uint16.tif'
135+
#cnn_path = '../nfk_artifacts/data/missing-wedge/input_irt/recon_theta_pm60_spac_2_uint16.tif'
136+
gt_path = './plot2/full_img_uint8_L_1256_W_780/mk_L291_tv_000163_L_1286_W_780_2_uint8.png'
137+
cnn_path = './plot2/full_img_uint8_L_1256_W_780/recon_theta_pm60_spac_2_L_1286_W_780_2_uint8.png' # missing_wedge
138+
dist_path = './plot2/full_img_uint8_L_1256_W_780/recon_theta_deg10_spac_0.5_L_1286_W_780_2_uint8.png' # distortion
139+
out_path = './plot2/full_fig/Ncomp_'+ str(Ncomp)+ '_uint8/'
140+
141+
if not os.path.isdir(out_path): os.makedirs(out_path, )
142+
if not crop_fig: dist_img = io_func.imageio_imread(dist_path)
143+
144+
gt_img = io_func.imageio_imread(gt_path)
145+
cnn_img = io_func.imageio_imread(cnn_path)
146+
h, w = gt_img.shape
147+
r = int(h/2)
148+
149+
if Ncomp == 3:
150+
fs_c1 = int(0.1*r)
151+
fs_c2 = int(0.25*r)
152+
153+
cnn_c1, cnn_c2, cnn_c3, cnn_c_stack = get3_comp_of_img(cnn_img, fs_c1, fs_c2, r)
154+
gt_c1, gt_c2, gt_c3, gt_c_stack = get3_comp_of_img(gt_img, fs_c1, fs_c2, r)
155+
156+
cnn_fft_c1, cnn_fft_c2, cnn_fft_c3, cnn_fft_c_stack = get3_fft_bands_of_img(cnn_img, fs_c1, fs_c2, r)
157+
158+
if plot_fig:
159+
pf.multi2dplots(1, 4, cnn_c_stack, axis=0, passed_fig_att={'colorbar': False})
160+
pf.multi2dplots(1, 4, cnn_fft_c_stack, axis=0, passed_fig_att={'colorbar': False})
161+
pf.plot2dlayers(gt_img)
162+
163+
if save_fig:
164+
if not os.path.isdir(out_path): os.makedirs(out_path, exist_ok=True)
165+
# cnn part
166+
cnn_c1 = utils.normalize_data_ab(0, 255, cnn_c1).astype('uint8')
167+
cnn_c2 = utils.normalize_data_ab(0, 255, cnn_c2).astype('uint8')
168+
cnn_c3 = utils.normalize_data_ab(0, 255, cnn_c3).astype('uint8')
169+
170+
#gt parts
171+
gt_c1 = utils.normalize_data_ab(0, 255, gt_c1).astype('uint8')
172+
gt_c2 = utils.normalize_data_ab(0, 255, gt_c2).astype('uint8')
173+
gt_c3 = utils.normalize_data_ab(0, 255, gt_c3).astype('uint8')
174+
#cnn fft parts
175+
cnn_fft_c1 = utils.normalize_data_ab(0, 255, cnn_fft_c1).astype('uint8')
176+
cnn_fft_c2 = utils.normalize_data_ab(0, 255, cnn_fft_c2).astype('uint8')
177+
cnn_fft_c3 = utils.normalize_data_ab(0, 255, cnn_fft_c3).astype('uint8')
178+
cnn_fft = utils.normalize_data_ab(0, 255, cnn_fft_c_stack[0]).astype('uint8')
179+
io_func.imsave(cnn_c1, path=out_path + 'cnn_c1.png', svtype='original')
180+
io_func.imsave(cnn_c2, path=out_path + 'cnn_c2.png', svtype='original')
181+
io_func.imsave(cnn_c3, path=out_path + 'cnn_c3.png', svtype='original')
182+
183+
io_func.imsave(gt_c1, path=out_path + 'gt_c1.png', svtype='original')
184+
io_func.imsave(gt_c2, path=out_path + 'gt_c2.png', svtype='original')
185+
io_func.imsave(gt_c3, path=out_path + 'gt_c3.png', svtype='original')
186+
187+
io_func.imsave(cnn_fft_c1, path=out_path + 'cnn_fft_c1.png', svtype='original')
188+
io_func.imsave(cnn_fft_c2, path=out_path + 'cnn_fft_c2.png', svtype='original')
189+
io_func.imsave(cnn_fft_c3, path=out_path + 'cnn_fft_c3.png', svtype='original')
190+
io_func.imsave(cnn_fft, path=out_path + 'cnn_fft.png', svtype='original')
191+
192+
elif Ncomp==5:
193+
fs_c1 = int(0.1*r)
194+
fs_c2 = int(0.25*r)
195+
fs_c3 = int(0.5*r)
196+
fs_c4 = int(0.75*r)
197+
198+
cnn_c1, cnn_c2, cnn_c3, cnn_c4, cnn_c5, cnn_c_stack = get5_comp_of_img(cnn_img, fs_c1, fs_c2, fs_c3, fs_c4, r) # missing-wedge-img/srgan-patch
199+
gt_c1, gt_c2, gt_c3, gt_c4, gt_c5, gt_c_stack = get5_comp_of_img(gt_img, fs_c1, fs_c2, fs_c3, fs_c4, r)
200+
cnn_fft_c1, cnn_fft_c2, cnn_fft_c3, cnn_fft_c4, cnn_fft_c5, cnn_fft_c_stack = get5_fft_bands_of_img(cnn_img, fs_c1, fs_c2, fs_c3, fs_c4, r)
201+
202+
if not crop_fig:
203+
dist_c1, dist_c2, dist_c3, dist_c4, dist_c5, dist_c_stack = get5_comp_of_img(dist_img, fs_c1, fs_c2, fs_c3, fs_c4, r)
204+
205+
if plot_fig:
206+
if crop_fig:
207+
# the three rows correspond to FFT of SRGAN, SRGAN and FBP
208+
band_stacks = np.stack((cnn_fft_c_stack.reshape(6, h, w), cnn_c_stack.reshape(6, h, w), gt_c_stack.reshape(6, h, w)), axis=0)
209+
print('shape of the subplots:', (band_stacks.reshape(18, h, w)).shape)
210+
pf.multi2dplots(3, 6, band_stacks.reshape(18, h, w), axis=0, passed_fig_att={'colorbar': False, 'figsize': [8, 6]})#, 'out_path': out_path + 'all_bands.png'})
211+
else:
212+
band_stacks = np.stack((cnn_fft_c_stack.reshape(6, h, w), cnn_c_stack.reshape(6, h, w), dist_c_stack.reshape(6, h, w), gt_c_stack.reshape(6, h, w)), axis=0)
213+
print('shape of the subplots:', (band_stacks.reshape(24, h, w)).shape)
214+
pf.multi2dplots(4, 6, band_stacks.reshape(24, h, w), axis=0, passed_fig_att={'colorbar': False, 'figsize': [8, 6]})#, 'out_path': out_path + 'all_bands.png'})
215+
216+
if save_fig:
217+
if not os.path.isdir(out_path): os.makedirs(out_path, exist_ok=True)
218+
# cnn part
219+
cnn_c1 = utils.normalize_data_ab(0, 255, cnn_c1).astype('uint8')
220+
cnn_c2 = utils.normalize_data_ab(0, 255, cnn_c2).astype('uint8')
221+
cnn_c3 = utils.normalize_data_ab(0, 255, cnn_c3).astype('uint8')
222+
cnn_c4 = utils.normalize_data_ab(0, 255, cnn_c4).astype('uint8')
223+
cnn_c5 = utils.normalize_data_ab(0, 255, cnn_c5).astype('uint8')
224+
225+
#gt parts
226+
gt_c1 = utils.normalize_data_ab(0, 255, gt_c1).astype('uint8')
227+
gt_c2 = utils.normalize_data_ab(0, 255, gt_c2).astype('uint8')
228+
gt_c3 = utils.normalize_data_ab(0, 255, gt_c3).astype('uint8')
229+
gt_c4 = utils.normalize_data_ab(0, 255, gt_c4).astype('uint8')
230+
gt_c5 = utils.normalize_data_ab(0, 255, gt_c5).astype('uint8')
231+
232+
#cnn fft parts
233+
cnn_fft_c1 = utils.normalize_data_ab(0, 255, cnn_fft_c1).astype('uint8')
234+
cnn_fft_c2 = utils.normalize_data_ab(0, 255, cnn_fft_c2).astype('uint8')
235+
cnn_fft_c3 = utils.normalize_data_ab(0, 255, cnn_fft_c3).astype('uint8')
236+
cnn_fft_c4 = utils.normalize_data_ab(0, 255, cnn_fft_c4).astype('uint8')
237+
cnn_fft_c5 = utils.normalize_data_ab(0, 255, cnn_fft_c5).astype('uint8')
238+
cnn_fft = utils.normalize_data_ab(0, 255, cnn_fft_c_stack[0]).astype('uint8')
239+
240+
io_func.imsave(cnn_c1, path=out_path + 'cnn_c1.png', svtype='original')
241+
io_func.imsave(cnn_c2, path=out_path + 'cnn_c2.png', svtype='original')
242+
io_func.imsave(cnn_c3, path=out_path + 'cnn_c3.png', svtype='original')
243+
io_func.imsave(cnn_c4, path=out_path + 'cnn_c4.png', svtype='original')
244+
io_func.imsave(cnn_c5, path=out_path + 'cnn_c5.png', svtype='original')
245+
246+
io_func.imsave(gt_c1, path=out_path + 'gt_c1.png', svtype='original')
247+
io_func.imsave(gt_c2, path=out_path + 'gt_c2.png', svtype='original')
248+
io_func.imsave(gt_c3, path=out_path + 'gt_c3.png', svtype='original')
249+
io_func.imsave(gt_c4, path=out_path + 'gt_c4.png', svtype='original')
250+
io_func.imsave(gt_c5, path=out_path + 'gt_c5.png', svtype='original')
251+
252+
io_func.imsave(cnn_fft_c1, path=out_path + 'cnn_fft_c1.png', svtype='original')
253+
io_func.imsave(cnn_fft_c2, path=out_path + 'cnn_fft_c2.png', svtype='original')
254+
io_func.imsave(cnn_fft_c3, path=out_path + 'cnn_fft_c3.png', svtype='original')
255+
io_func.imsave(cnn_fft_c4, path=out_path + 'cnn_fft_c4.png', svtype='original')
256+
io_func.imsave(cnn_fft_c5, path=out_path + 'cnn_fft_c5.png', svtype='original')
257+
io_func.imsave(cnn_fft, path=out_path + 'cnn_fft.png', svtype='original')
258+
else:
259+
sys.exit('the only available options are 3 and 5')
260+

‎paper_plots/crop_banded/crop_img_uint16/gt_000069.raw‎

15.8 KB
Binary file not shown.

‎paper_plots/crop_banded/crop_img_uint16/srgan_000069.raw‎

15.8 KB
Binary file not shown.

‎paper_plots/plot2/crop_img_uint8_L_50_w_400/gt_000069.png‎ renamed to ‎paper_plots/crop_banded/crop_img_uint8_L_50_W_400/gt_000069.png‎

File renamed without changes.

‎paper_plots/plot2/crop_img_uint8_L_50_w_400/srgan_000069.png‎ renamed to ‎paper_plots/crop_banded/crop_img_uint8_L_50_W_400/srgan_000069.png‎

File renamed without changes.

‎paper_plots/git_illustration2.png‎

1.27 MB
Loading

‎paper_plots/git_illustration_conv_artifact_banded_plots.png‎

227 KB
Loading

‎paper_plots/git_illustration_distortion.png‎

178 KB
Loading

‎paper_plots/git_illustration_indentation_all_bands.png‎

355 KB
Loading

‎paper_plots/git_illustration_missing_wedge.png‎

175 KB
Loading

0 commit comments

Comments
 (0)