-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathbrushstroke_LDDP.py
192 lines (162 loc) · 7.32 KB
/
brushstroke_LDDP.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
import numpy as np
from DDP import *
import cv2
import pdb
if __name__ == '__main__':
SAMPLE_IMAGE_PATH = 'brushstroke_sample_004.png'
BRUSHSTROKES_NUM = 30
BRUSHSTROKES_THICKNESS = 15
LEARNING_RATE = 0.2
SAVE_DIR = 'save'
SAVE_IDENTIFIER = 'brushstroke_LDDP'
SHOULD_SAVE_DRAW = True
SHOULD_SAVE_FRAMES = True
SHOULD_SAVE_VIDEO = False
image_desired = cv2.imread(SAMPLE_IMAGE_PATH)
image_desired = cv2.cvtColor(image_desired, cv2.COLOR_BGR2GRAY)
ret_info, image_desired = cv2.threshold(image_desired, 127, 255, cv2.THRESH_BINARY)
#cv2.imshow('image_desired', image_desired)
video_writer = None
if SHOULD_SAVE_VIDEO:
fourcc = cv2.VideoWriter_fourcc(*'H264')
save_video_path = SAVE_DIR + '/' + SAVE_IDENTIFIER + '.avi'
video_writer = cv2.VideoWriter(save_video_path, fourcc, 5.0, image_desired.shape, isColor=False)
# nodes
nodes = []
for i in range(BRUSHSTROKES_NUM + 1):
node = DynamicsSystemNode('picture_' + str(i))
if (i < BRUSHSTROKES_NUM):
node.next_primitive_name = 'draw_' + str(i)
bound = image_desired.shape
init_p1 = np.random.sample(2) * np.array([bound[0], bound[1]])
init_p2 = init_p1 + np.array([5., 5.])
init_brushstroke = np.array([init_p1[0], init_p1[1], init_p2[0], init_p2[1]])
node.ssa.updateActionElement('brushstroke_' + str(i), init_brushstroke)
node.ssa.updateSelection(True)
if (i > 0):
node.prev_primitive_name = 'draw_' + str(i-1)
nodes.append(node)
nodes[0].ssa.updateInfo('image_draw', np.zeros(image_desired.shape, np.uint8))
# dynamics models
def simulate_brushstroke(image_draw, brushstroke):
image_brushstroke = np.zeros(image_desired.shape, np.uint8)
bounded_brushstroke = np.zeros(brushstroke.shape[0], np.int)
bounded_brushstroke[0] = max(0, min(image_desired.shape[1], np.round(brushstroke[0])))
bounded_brushstroke[1] = max(0, min(image_desired.shape[0], np.round(brushstroke[1])))
bounded_brushstroke[2] = max(0, min(image_desired.shape[1], np.round(brushstroke[2])))
bounded_brushstroke[3] = max(0, min(image_desired.shape[0], np.round(brushstroke[3])))
cv2.line(image_brushstroke, \
(bounded_brushstroke[0], bounded_brushstroke[1]), \
(bounded_brushstroke[2], bounded_brushstroke[3]), \
255, \
thickness=BRUSHSTROKES_THICKNESS)
#cv2.imshow('image_brushstroke', image_brushstroke)
image_draw_next = cv2.bitwise_or(image_draw, image_brushstroke)
#cv2.imshow('image_draw_next', image_draw_next)
image_increment = cv2.bitwise_xor(image_draw_next, image_draw)
#cv2.imshow('image_increment', image_increment)
image_matched = cv2.bitwise_and(image_desired, image_increment)
image_mismatched = cv2.bitwise_and(cv2.bitwise_xor(image_desired, image_increment), image_increment)
reward_positive = cv2.countNonZero(image_matched)
reward_negative = cv2.countNonZero(image_mismatched)
reward = reward_positive - reward_negative
#cv2.imshow('image_matched', image_matched)
#cv2.imshow('image_mismatched', image_mismatched)
return (image_draw_next, image_increment, reward)
dynamics_models = []
for i in range(BRUSHSTROKES_NUM):
dynamics = DynamicsSystemEdgeDynamicsModel()
def make_dynamics_func (i):
def dynamics_func (ssa):
next_ssa = ssa.copy()
image_draw = ssa.retriveInfo('image_draw')
brushstroke = ssa.retrive('brushstroke_' + str(i)) # p1_x, p1_y, p2_x, p2_y
image_draw_next, image_increment, reward = simulate_brushstroke(image_draw, brushstroke)
next_ssa.updateInfo('image_draw', image_draw_next)
next_ssa.updateInfo('image_increment', image_increment)
next_ssa.updateInfo('reward', reward)
reward_d = [0, 0, 0, 0]
for brushstroke_i in range(brushstroke.shape[0]):
di = np.zeros(brushstroke.shape[0])
di[brushstroke_i] += 6 # pixel precision
brushstroke_di = brushstroke + di
image_draw_next, image_increment, reward_di = simulate_brushstroke(image_draw, brushstroke_di)
reward_d[brushstroke_i] = (reward_di - reward) / di[brushstroke_i]
next_ssa.updateInfo('reward_d', reward_d)
return next_ssa
return dynamics_func
dynamics.dynamics_func = make_dynamics_func(i)
def make_dynamics_dfunc (i):
def dynamics_dfunc (ssa):
next_ssa = make_dynamics_func(i)(ssa)
d = {}
return d
return dynamics_dfunc
dynamics.dynamics_dfunc = make_dynamics_dfunc(i)
def make_reward_func (i):
def reward_func (next_ssa):
reward = next_ssa.retriveInfo('reward')
return reward
return reward_func
dynamics.reward_func = make_reward_func(i)
def make_reward_dfunc (i):
def reward_dfunc (next_ssa):
d = {}
for k in next_ssa.keys():
if k == 'selection':
d[k] = np.zeros((1))
else:
d[k] = np.zeros((next_ssa.retrive(k).shape[0]))
for brushstroke_i in range(next_ssa.retrive('brushstroke_' + str(i)).shape[0]):
d['brushstroke_' + str(i)][brushstroke_i] = next_ssa.retriveInfo('reward_d')[brushstroke_i]
return d
return reward_dfunc
dynamics.reward_dfunc = make_reward_dfunc(i)
dynamics_models.append(dynamics)
# dynamics system
dynamics_system = LinearDynamicsSystem()
for i in range(len(nodes)):
dynamics_system.updateNode('picture_' + str(i), nodes[i])
dynamics_system.updateRootNodeName('picture_0')
for i in range(len(dynamics_models)):
dynamics_system.updateDynamics('draw_' + str(i), \
'picture_' + str(i), \
dynamics_models[i], \
'picture_' + str(i+1), \
alpha=LEARNING_RATE)
# execute
for i in range(100):
J, a_err = dynamics_system.optimizeActionsOnce()
print('Round: {}, Value: {}, Error: {}'.format(i, J, a_err))
last_node_name = 'picture_' + str(len(nodes) - 1)
last_image_draw = nodes[len(nodes) - 1].ssa.retriveInfo('image_draw')
image_draw_fit = cv2.bitwise_xor(last_image_draw, image_desired)
if SHOULD_SAVE_VIDEO:
video_writer.write(image_draw_fit)
if SHOULD_SAVE_FRAMES:
save_frame_index = str(i)
while len(save_frame_index) < 4:
save_frame_index = '0' + save_frame_index
save_frame_path = SAVE_DIR + '/' + SAVE_IDENTIFIER + '_' + save_frame_index + '.jpg'
cv2.imwrite(save_frame_path, image_draw_fit)
cv2.imshow(last_node_name, image_draw_fit)
cv2.waitKey(25)
if a_err < 0.5:
break
if SHOULD_SAVE_VIDEO:
video_writer.release()
image_draw_final = np.zeros(image_desired.shape, np.uint8)
for i in range(len(nodes) - 1):
brushstroke = nodes[i].ssa.retrive('brushstroke_' + str(i))
print('brushstroke #{}: ({}, {}), ({}, {})'.format(
i,
np.round(brushstroke[0], 2), np.round(brushstroke[1], 2),
np.round(brushstroke[2], 2), np.round(brushstroke[3], 2)
))
image_draw_final, image_increment, reward = simulate_brushstroke(image_draw_final, brushstroke)
cv2.imshow('image_draw_final_fit', cv2.bitwise_xor(image_draw_final, image_desired))
cv2.imshow('image_draw_final', image_draw_final)
if SHOULD_SAVE_DRAW:
save_draw_path = SAVE_DIR + '/' + SAVE_IDENTIFIER + '_draw.jpg'
cv2.imwrite(save_draw_path, image_draw_final)
cv2.waitKey()