-
Notifications
You must be signed in to change notification settings - Fork 5
/
Oil-Painting.py
217 lines (177 loc) · 11 KB
/
Oil-Painting.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
import numpy as np
import cv2, random, time, os, argparse
from simulate_RGB import *
from drawpatch import *
from ETF.edge_tangent_flow import *
from quicksort import *
from voronoi_sampler import K_Means_Sampler
from search_and_render import *
if __name__ == '__main__':
default = {
# config parameters (user control)
"image":"./input/S1.jpg", # input image filepath
"brush":"./brush/brush-0.png", # brush template
"p_max": 4, # the reciprocal of the Maximum Sampling Rate, use 4, 9, 16, 25, 36
"seed": 0, # np.random.seed()
"force": True, # force recomputation of the anchor Map
"SSAA" : 8, # Super-Sampling Anti-Aliasing
"freq" : 100, # save one frame every（freq) strokes drawn
"stroke_order_type": 0, # use 0 for the default size order, use 1 for random order
# default parameters (don't change)
"padding": 5, # padding
"n_iter": 15, # K-means iteration
"k_size": 5, # Sobel and Mean Filter size
"figsize": 6, # anchor map figure size
"pointsize": (8.0, 8.0), # point (mix,max) size for the anchor map
"ratio" : 3, # max_length/max_width
"threshold_hsv": (30,None,15), # threshold for hsv color space during searching
"kernel_radius" : 5, # ETF kernel_radius
"ETF_iter" : 15, # ETF iteration number
"background_dir" : None, # for ETF
}
description = "Im2Oil: Stroke-Based Oil Painting Rendering with Linearly Controllable Fineness Via Adaptive Sampling"
parser = argparse.ArgumentParser(description=description)
parser.add_argument('--f', type=str, default=default["image"],
help='input image path')
parser.add_argument('--b', type=str, default=default["brush"],
help='brush template path')
parser.add_argument('--p', type=int, default=default["p_max"],
help='maximum sampling rate')
parser.add_argument('--s', type=int, default=default["seed"],
help='np.random.seed()')
parser.add_argument('--force', action='store_true', default=default["force"],
help='Force recomputation of the Anchor Map')
parser.add_argument('--SSAA', type=int, default=default["SSAA"],
help='Super-Sampling Anti-Aliasing')
parser.add_argument('--freq', type=int, default=default["freq"],
help='save one frame every (freq) strokes drawn')
parser.add_argument('--order', type=int, default=default["stroke_order_type"],
help='0 for default size order, 1 for random order')
args = parser.parse_args()
# auto parameters (before SSAA)
p_max = 1.0/args.p # maximum sampling rate
p_min = p_max/100 # minimum sampling rate
ratio = default["ratio"] # max_length/max_width
max_width = np.sqrt(1/p_min) # maximum stroke width
min_width = np.sqrt(1/p_max)-1 # minimum stroke width
max_length = int(ratio * max_width) # maximum stroke length
min_length = ratio * min_width # minimum stroke length
padding = default["padding"] # padding
####### make directory #######
if 1:
filename = os.path.basename(args.f)
print("filename:", filename)
filename = filename.split('.')[0]
# point_path = './output/'+filename+"-"+str(p_max)+"/"+filename+'-'+str(point_num)+".npy"
brush_path = args.b
brush = cv2.imread(brush_path, cv2.IMREAD_GRAYSCALE)
output_path = './output/'+filename+'-p-'+str(args.p)
if not os.path.exists(output_path):
os.makedirs(output_path)
os.makedirs(output_path+"/anchor")
os.makedirs(output_path+"/stroke")
os.makedirs(output_path+"/process")
######## K-means ########
np.random.seed(args.s)
point_num, density, gradient_magnitude, point_path = \
K_Means_Sampler(output_dir=output_path+"/anchor", filename=args.f, p_max=p_max, p_min=p_min,
border_copy=padding, k_size=default["k_size"], n_iter=default["n_iter"], figsize=default["figsize"], pointsize=default["pointsize"],
display=False, force=args.force, save=True)
####### save input #######
if 1:
input_bgr = cv2.imread(args.f, cv2.IMREAD_COLOR) # bgr输入
cv2.imwrite(output_path + "/input_bgr.png", input_bgr)
input_bgr = cv2.copyMakeBorder(input_bgr, padding, padding, padding, padding, cv2.BORDER_REPLICATE)
input_hsv = cv2.cvtColor(input_bgr, cv2.COLOR_BGR2HSV) # hsv输入
input_gray = cv2.cvtColor(input_bgr, cv2.COLOR_BGR2GRAY) # gray输入
(H0,W0) = input_gray.shape
####### ETF #######
if 1:
time_start=time.time()
ETF_filter = ETF(img=input_gray, output_path=output_path+'/mask',
kernel_radius=kernel_radius, iter_time=default["ETF_iter"], background_dir=default["background_dir"])
angle = ETF_filter.forward().numpy()
angle_hatch = angle+90
angle_hatch[angle_hatch>90] -= 180
print('ETF Filtering time:', int(time.time()-time_start),"seconds")
print('ETF done')
############ Search patch ##########
if 1:
time_start=time.time() # time
points = np.load(point_path) # stipple coordinates
patch_sequence = Search_Stroke(points, density, input_gray, input_hsv, gradient_magnitude, ratio,
angle, angle_hatch, min_width, min_length, max_width, max_length, default["threshold_hsv"])
print('Stoke Searching time', int(time.time()-time_start),"seconds")
print('Stoke number', len(patch_sequence))
############ Stroke Order ##########
if args.order == 1:
random.shuffle(patch_sequence)
elif args.order == 0:
random.shuffle(patch_sequence)
quickSort(patch_sequence,0,len(patch_sequence)-1)
############ Render Stroke ##########
if 1:
SSAA = args.SSAA
freq = args.freq
time_start=time.time() # time
wihte = Gassian_HSV((H0*SSAA-2*padding*SSAA,W0*SSAA-2*padding*SSAA,3)) # padding
cv2.imwrite(output_path + "/process/{0:04d}.png".format(0), cv2.cvtColor(wihte, cv2.COLOR_HSV2BGR))
Canvas, Mask = Render_Stroke(brush, patch_sequence, input_gray, output_path, max_length, SSAA=SSAA, BORDERCOPY=padding, FREQ=freq, save=True)
print('Stoke Rendering time', int(time.time()-time_start),"seconds")
print('Stoke number', len(patch_sequence))
result = Canvas[max_length*SSAA:-max_length*SSAA,max_length*SSAA:-max_length*SSAA]
cv2.imwrite(output_path + "/Oil_drawing.png", cv2.cvtColor(result, cv2.COLOR_HSV2BGR))
########### Pad Blank Area ###########
if 1:
mask = Mask
mask[ max_length*SSAA+padding*SSAA-1,:] = 1
mask[-max_length*SSAA-padding*SSAA,:] = 1
mask[:, max_length*SSAA+padding*SSAA-1] = 1
mask[:,-max_length*SSAA-padding*SSAA] = 1
# cv2.imshow('mask', np.uint8(mask*255))
# cv2.waitKey(0)
while(1):
result = cv2.imread(output_path + "/Oil_drawing.png", cv2.IMREAD_COLOR) # label输入
mask_cut = mask[max_length*SSAA:-max_length*SSAA,max_length*SSAA:-max_length*SSAA]
cv2.imwrite(output_path + "/mask.png", mask_cut.astype("uint8")*255)
connect_num, labels, stats, centroids = cv2.connectedComponentsWithStats(255-mask_cut.astype("uint8")*255, connectivity=8)
Points = []
for i in range(centroids.shape[0]):
p = centroids[i]
if p[0] >= padding*SSAA and p[1] >= padding*SSAA and p[0] < result.shape[1]-padding*SSAA and p[1] < result.shape[0]-padding*SSAA and stats[i][4]>0 and stats[i][4]<result.shape[0]*result.shape[1]/4:
p[0], p[1] = p[0]/SSAA, p[1]/SSAA
Points.append([p[0],result.shape[0]/SSAA-p[1]]) # x, y
Points = np.array(Points)
if Points.shape[0] == 0:
cv2.imwrite(output_path + "/Final_Result.png", result[padding*SSAA:-padding*SSAA,padding*SSAA:-padding*SSAA])
cv2.imwrite(output_path + "/process/Final_Result.png", result[padding*SSAA:-padding*SSAA,padding*SSAA:-padding*SSAA])
break
else:
for point in Points:
cv2.circle(result, (int(np.around(point[0]*SSAA)),int(np.around((result.shape[0]/SSAA-point[1])*SSAA))), 3, (0,0,255), 3)
cv2.imwrite(output_path + "/anchor.png", result)
#### Search ###
pad_sequence = Search_Stroke(np.array(Points), density, input_gray, input_hsv, gradient_magnitude, ratio,
angle, angle_hatch, min_width, min_length, max_width, max_length, default["threshold_hsv"])
if args.order == 0:
quickSort(pad_sequence,0,len(pad_sequence)-1)
### Pad ###
pad_canvas, pad_mask = Render_Stroke(brush, pad_sequence, input_gray, output_path, max_length, SSAA=SSAA, BORDERCOPY=padding, FREQ=freq, save=False)
pad_canvas_cut = pad_canvas[max_length*SSAA:-max_length*SSAA,max_length*SSAA:-max_length*SSAA]
pad_canvas_cut = cv2.cvtColor(pad_canvas_cut, cv2.COLOR_HSV2BGR)
for point in Points:
cv2.circle(pad_canvas_cut, (int(np.around(point[0]*SSAA)),int(np.around((result.shape[0]/SSAA-point[1])*SSAA))), 3, (0,0,255), 3)
cv2.imwrite(output_path + "/pad_canvas_cut.png", pad_canvas_cut)
Oil_drawing = cv2.imread(output_path + "/Oil_drawing.png", cv2.IMREAD_COLOR) # label输入
Oil_drawing = cv2.cvtColor(Oil_drawing, cv2.COLOR_BGR2HSV)
pad_canvas_cut = pad_canvas[max_length*SSAA:-max_length*SSAA,max_length*SSAA:-max_length*SSAA]
m = pad_mask*(1-mask)
m = m[max_length*SSAA:-max_length*SSAA,max_length*SSAA:-max_length*SSAA,np.newaxis]
Oil_drawing = np.uint8(m*pad_canvas_cut + (1-m)*Oil_drawing)
Oil_drawing = cv2.cvtColor(Oil_drawing, cv2.COLOR_HSV2BGR)
cv2.imwrite(output_path + "/Oil_drawing.png", Oil_drawing)
mask += pad_mask
mask[mask>0]=1
# result[:,:,2] = np.uint8(np.around(np.clip(result[:,:,2].astype("float32")/0.85,0,255)))
min_length += ratio
min_width += 1