-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathrun_object_sketching.py
187 lines (159 loc) · 7.57 KB
/
run_object_sketching.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
import sys
import warnings
warnings.filterwarnings('ignore')
warnings.simplefilter('ignore')
import argparse
import multiprocessing as mp
import os
import subprocess as sp
from shutil import copyfile
import numpy as np
import torch
from IPython.display import Image as Image_colab
from IPython.display import display, SVG, clear_output
from ipywidgets import IntSlider, Output, IntProgress, Button
import time
parser = argparse.ArgumentParser()
parser.add_argument("--target_file", type=str,
help="target image file, located in <target_images>")
parser.add_argument("--output_name", type=str,
help="target image file, located in <target_images>")
parser.add_argument("--num_strokes", type=int, default=16,
help="number of strokes used to generate the sketch, this defines the level of abstraction.")
parser.add_argument("--num_iter", type=int, default=1001,
help="number of iterations")
parser.add_argument("--num_segments", type=int, default=5)
parser.add_argument("--fix_scale", type=int, default=0,
help="if the target image is not squared, it is recommended to fix the scale")
parser.add_argument("--mask_object", type=int, default=0,
help="if the target image contains background, it's better to mask it out")
parser.add_argument("--num_sketches", type=int, default=3,
help="it is recommended to draw 3 sketches and automatically chose the best one")
parser.add_argument("--multiprocess", type=int, default=0,
help="recommended to use multiprocess if your computer has enough memory")
parser.add_argument('--train_with_diffusion', action='store_true')
parser.add_argument('--control', action='store_true')
parser.add_argument('--sd_version', type=str, default='1.5', choices=['1.5', '2.0', '2.1'], help="stable diffusion version")
parser.add_argument('--hf_key', type=str, default=None, help="hugging face Stable diffusion model key")
parser.add_argument('--fp16', action='store_true', help="use float16 for training")
parser.add_argument('--vram_O', action='store_true', help="optimization for low VRAM usage")
parser.add_argument('--text_prompt', type=str, default=None)
parser.add_argument('--text_negative_prompt', type=str, default='')
parser.add_argument("--attention_init", type=int, default=0,
help="if True, use the attention heads of Dino model to set the location of the initial strokes")
parser.add_argument("--sketches_edit", type=str, default='none')
parser.add_argument('--bbox', type=str, default='none')
parser.add_argument('--init_point', type=str, default='none')
parser.add_argument('-colab', action='store_true')
parser.add_argument('-cpu', action='store_true')
parser.add_argument('-display', action='store_true')
parser.add_argument("--points_init", type=str, default='none')
args = parser.parse_args()
multiprocess = not args.colab and args.num_sketches > 1 and args.multiprocess
abs_path = '/home/SketchDreamer'
target = f"{abs_path}/input/{args.target_file}"
if not os.path.isfile(f"{abs_path}/U2Net_/saved_models/u2net.pth"):
sp.run(["gdown", "https://drive.google.com/uc?id=1ao1ovG1Qtx4b7EoskHXmi2E9rp5CHLcZ",
"-O", "U2Net_/saved_models/"])
test_name = args.output_name
output_dir = f"{abs_path}/output/{test_name}/"
if not os.path.exists(output_dir):
os.makedirs(output_dir)
num_iter = args.num_iter
save_interval = 50
use_gpu = not args.cpu
if not torch.cuda.is_available():
use_gpu = False
print("CUDA is not configured with GPU, running with CPU instead.")
print("Note that this will be very slow, it is recommended to use colab.")
if args.colab:
print("=" * 50)
print(f"Processing [{args.target_file}] ...")
if args.colab or args.display:
img_ = Image_colab(target)
display(img_)
print(f"GPU: {use_gpu}, {torch.cuda.current_device()}")
print(f"Results will be saved to \n[{output_dir}] ...")
print("=" * 50)
seeds = list(range(0, args.num_sketches * 1, 1))
exit_codes = []
manager = mp.Manager()
losses_all = manager.dict()
def run(seed, wandb_name):
exit_code = sp.run(["python", f"{abs_path}/painterly_rendering.py", target,
"--num_paths", str(args.num_strokes),
"--output_dir", output_dir,
"--wandb_name", wandb_name,
"--num_iter", str(num_iter),
"--num_segments", str(args.num_segments),
"--save_interval", str(save_interval),
"--seed", str(seed),
"--use_gpu", str(int(use_gpu)),
"--fix_scale", str(args.fix_scale),
"--mask_object", str(args.mask_object),
"--mask_object_attention", str(args.mask_object),
"--display_logs", str(int(args.colab)),
"--display", str(int(args.display)),
"--points_init", str(args.points_init),
"--sketches_edit", str(args.sketches_edit),
"--attention_init", str(args.attention_init),
"--train_with_diffusion", str(int(args.train_with_diffusion)),
"--control", str(int(args.control)),
"--sd_version", str(args.sd_version),
"--hf_key", str(args.hf_key),
"--fp16", str(args.fp16),
"--vram_O", str(args.vram_O),
"--text_prompt", str(args.text_prompt),
"--text_negative_prompt", str(args.text_negative_prompt),
'--bbox', str(args.bbox),
'--init_point', str(args.init_point),
])
if exit_code.returncode:
sys.exit(1)
config = np.load(f"{output_dir}/{wandb_name}/config.npy",
allow_pickle=True)[()]
loss_eval = np.array(config['loss_eval'])
inds = np.argsort(loss_eval)
losses_all[wandb_name] = loss_eval[inds][0]
def display_(seed, wandb_name):
path_to_svg = f"{output_dir}/{wandb_name}/svg_logs/"
intervals_ = list(range(0, num_iter, save_interval))
filename = f"svg_iter0.svg"
display(IntSlider())
out = Output()
display(out)
for i in intervals_:
filename = f"svg_iter{i}.svg"
not_exist = True
while not_exist:
not_exist = not os.path.isfile(f"{path_to_svg}/{filename}")
continue
with out:
clear_output()
print("")
display(IntProgress(
value=i,
min=0,
max=num_iter,
description='Processing:',
bar_style='info', # 'success', 'info', 'warning', 'danger' or ''
style={'bar_color': 'maroon'},
orientation='horizontal'
))
display(SVG(f"{path_to_svg}/svg_iter{i}.svg"))
if multiprocess:
ncpus = 10
P = mp.Pool(ncpus)
for seed in seeds:
wandb_name = f"{test_name}_{args.num_strokes}strokes_seed{seed}"
if multiprocess:
P.apply_async(run, (seed, wandb_name))
else:
run(seed, wandb_name)
if args.display:
time.sleep(10)
P.apply_async(display_, (0, f"{test_name}_{args.num_strokes}strokes_seed0"))
if multiprocess:
P.close()
P.join() # start processes
sorted_final = dict(sorted(losses_all.items(), key=lambda item: item[1]))