-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathsketck_rnn_sample.py
210 lines (174 loc) · 6.81 KB
/
sketck_rnn_sample.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
# import the required libraries
import numpy as np
import time
import random
import codecs
import collections
import os
import math
import json
import tensorflow as tf
from six.moves import xrange
# libraries required for visualisation:
from IPython.display import SVG, display
import PIL
from PIL import Image
import matplotlib.pyplot as plt
np.set_printoptions(precision=8, edgeitems=6, linewidth=200, suppress=True)
from magenta.models.sketch_rnn.sketch_rnn_train import *
from magenta.models.sketch_rnn.model import *
from magenta.models.sketch_rnn.utils import *
from magenta.models.sketch_rnn.rnn import *
import svgwrite
from six.moves import xrange
from IPython.display import SVG, display
def draw_strokes(data, factor=0.2, svg_filename = '/tmp/sketch_rnn/svg/sample.svg'):
tf.gfile.MakeDirs(os.path.dirname(svg_filename))
min_x, max_x, min_y, max_y = get_bounds(data, factor)
dims = (50 + max_x - min_x, 50 + max_y - min_y)
dwg = svgwrite.Drawing(svg_filename, size=dims)
dwg.add(dwg.rect(insert=(0, 0), size=dims,fill='white'))
lift_pen = 1
abs_x = 25 - min_x
abs_y = 25 - min_y
p = "M%s,%s " % (abs_x, abs_y)
command = "m"
for i in xrange(len(data)):
if (lift_pen == 1):
command = "m"
elif (command != "l"):
command = "l"
else:
command = ""
x = float(data[i,0])/factor
y = float(data[i,1])/factor
lift_pen = data[i, 2]
p += command+str(x)+","+str(y)+" "
the_color = "black"
stroke_width = 1
dwg.add(dwg.path(p).stroke(the_color,stroke_width).fill("none"))
dwg.save()
display(SVG(dwg.tostring()))
def make_grid_svg(s_list, grid_space=10.0, grid_space_x=16.0):
def get_start_and_end(x):
x = np.array(x)
x = x[:, 0:2]
x_start = x[0]
x_end = x.sum(axis=0)
x = x.cumsum(axis=0)
x_max = x.max(axis=0)
x_min = x.min(axis=0)
center_loc = (x_max+x_min)*0.5
return x_start-center_loc, x_end
x_pos = 0.0
y_pos = 0.0
result = [[x_pos, y_pos, 1]]
for sample in s_list:
s = sample[0]
grid_loc = sample[1]
grid_y = grid_loc[0]*grid_space+grid_space*0.5
grid_x = grid_loc[1]*grid_space_x+grid_space_x*0.5
start_loc, delta_pos = get_start_and_end(s)
loc_x = start_loc[0]
loc_y = start_loc[1]
new_x_pos = grid_x+loc_x
new_y_pos = grid_y+loc_y
result.append([new_x_pos-x_pos, new_y_pos-y_pos, 0])
result += s.tolist()
result[-1][2] = 1
x_pos = new_x_pos+delta_pos[0]
y_pos = new_y_pos+delta_pos[1]
return np.array(result)
def load_env_compatible(data_dir, model_dir):
"""Loads environment for inference mode, used in jupyter notebook."""
# modified https://github.com/tensorflow/magenta/blob/master/magenta/models/sketch_rnn/sketch_rnn_train.py
# to work with depreciated tf.HParams functionality
model_params = sketch_rnn_model.get_default_hparams()
with tf.gfile.Open(os.path.join(model_dir, 'model_config.json'), 'r') as f:
data = json.load(f)
fix_list = ['conditional', 'is_training', 'use_input_dropout', 'use_output_dropout', 'use_recurrent_dropout']
for fix in fix_list:
data[fix] = (data[fix] == 1)
model_params.parse_json(json.dumps(data))
return load_dataset(data_dir, model_params, inference_mode=True)
# data_dir = 'http://github.com/hardmaru/sketch-rnn-datasets/raw/master/aaron_sheep/'
data_dir = '/Users/yuhaomao/Downloads/sketch-rnn-datasets/aaron_sheep'
models_root_dir = '/tmp/sketch_rnn/models'
model_dir = '/Users/yuhaomao/Downloads/sketch_rnn/aaron_sheep/layer_norm'
# download_pretrained_models(models_root_dir=models_root_dir)
def load_model_compatible(model_dir):
"""Loads model for inference mode, used in jupyter notebook."""
# modified https://github.com/tensorflow/magenta/blob/master/magenta/models/sketch_rnn/sketch_rnn_train.py
# to work with depreciated tf.HParams functionality
model_params = sketch_rnn_model.get_default_hparams()
with tf.gfile.Open(os.path.join(model_dir, 'model_config.json'), 'r') as f:
data = json.load(f)
fix_list = ['conditional', 'is_training', 'use_input_dropout', 'use_output_dropout', 'use_recurrent_dropout']
for fix in fix_list:
data[fix] = (data[fix] == 1)
model_params.parse_json(json.dumps(data))
model_params.batch_size = 1 # only sample one at a time
eval_model_params = sketch_rnn_model.copy_hparams(model_params)
eval_model_params.use_input_dropout = 0
eval_model_params.use_recurrent_dropout = 0
eval_model_params.use_output_dropout = 0
eval_model_params.is_training = 0
sample_model_params = sketch_rnn_model.copy_hparams(eval_model_params)
sample_model_params.max_seq_len = 1 # sample one point at a time
return [model_params, eval_model_params, sample_model_params]
[train_set, valid_set, test_set, hps_model, eval_hps_model, sample_hps_model] = load_env_compatible(data_dir, model_dir)
"""
have some error
add "allow_pickle = True" at "sketch_rnn_train.py" line 133
data = np.load(six.BytesIO(response.content), encoding='latin1',allow_pickle= True)
==================
"""
# construct the sketch-rnn model here:
reset_graph()
model = Model(hps_model)
eval_model = Model(eval_hps_model, reuse=True)
sample_model = Model(sample_hps_model, reuse=True)
sess = tf.InteractiveSession()
sess.run(tf.global_variables_initializer())
# loads the weights from checkpoint into our model
load_checkpoint(sess, model_dir)
def encode(input_strokes):
strokes = to_big_strokes(input_strokes).tolist()
strokes.insert(0, [0, 0, 1, 0, 0])
seq_len = [len(input_strokes)]
draw_strokes(to_normal_strokes(np.array(strokes)))
return sess.run(eval_model.batch_z, feed_dict={eval_model.input_data: [strokes], eval_model.sequence_lengths: seq_len})[0]
def decode(z_input=None, draw_mode=True, temperature=0.1, factor=0.2):
z = None
if z_input is not None:
z = [z_input]
sample_strokes, m = sample(sess, sample_model, seq_len=eval_model.hps.max_seq_len, temperature=temperature, z=z)
strokes = to_normal_strokes(sample_strokes)
if draw_mode:
draw_strokes(strokes, factor)
return strokes
# get a sample drawing from the test set, and render it to .svg
stroke = test_set.random_sample()
draw_strokes(stroke)
z = encode(stroke)
_ = decode(z, temperature=0.8) # convert z back to drawing at temperature of 0.8
stroke_list = []
for i in range(10):
stroke_list.append([decode(z, draw_mode=False, temperature=0.1*i+0.1), [0, i]])
stroke_grid = make_grid_svg(stroke_list)
draw_strokes(stroke_grid)
# get a sample drawing from the test set, and render it to .svg
z0 = z
_ = decode(z0)
stroke = test_set.random_sample()
z1 = encode(stroke)
_ = decode(z1)
z_list = [] # interpolate spherically between z0 and z1
N = 10
for t in np.linspace(0, 1, N):
z_list.append(slerp(z0, z1, t))
reconstructions = []
for i in range(N):
reconstructions.append([decode(z_list[i], draw_mode=False), [0, i]])
stroke_grid = make_grid_svg(reconstructions)
draw_strokes(stroke_grid)