Spaces:
Runtime error
Runtime error
sunshineatnoon
commited on
Commit
·
1d90a68
1
Parent(s):
d4e058e
Add application file
Browse files
app.py
ADDED
@@ -0,0 +1,307 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import time
|
3 |
+
import json
|
4 |
+
import base64
|
5 |
+
import argparse
|
6 |
+
import importlib
|
7 |
+
from glob import glob
|
8 |
+
from PIL import Image
|
9 |
+
from imageio import imsave
|
10 |
+
|
11 |
+
import torch
|
12 |
+
import torchvision.utils as vutils
|
13 |
+
|
14 |
+
import sys
|
15 |
+
sys.path.append(".")
|
16 |
+
|
17 |
+
import numpy as np
|
18 |
+
from libs.test_base import TesterBase
|
19 |
+
from libs.utils import colorEncode, label2one_hot_torch
|
20 |
+
from tqdm import tqdm
|
21 |
+
from libs.options import BaseOptions
|
22 |
+
from skimage.segmentation import mark_boundaries
|
23 |
+
import torch.nn.functional as F
|
24 |
+
from libs.nnutils import poolfeat, upfeat
|
25 |
+
|
26 |
+
import streamlit as st
|
27 |
+
from skimage.segmentation import slic
|
28 |
+
import torchvision.transforms.functional as TF
|
29 |
+
import torchvision.transforms as transforms
|
30 |
+
from st_clickable_images import clickable_images
|
31 |
+
|
32 |
+
args = BaseOptions().gather_options()
|
33 |
+
if args.img_path is not None:
|
34 |
+
args.exp_name = os.path.join(args.exp_name, args.img_path.split('/')[-1].split('.')[0])
|
35 |
+
args.batch_size = 1
|
36 |
+
args.data_path = "/home/xli/DATA/BSR_processed/train"
|
37 |
+
args.label_path = "/home/xli/DATA/BSR/BSDS500/data/groundTruth"
|
38 |
+
args.device = torch.device("cpu")
|
39 |
+
args.nsamples = 500
|
40 |
+
args.out_dir = os.path.join('cachedir', args.exp_name)
|
41 |
+
os.makedirs(args.out_dir, exist_ok=True)
|
42 |
+
args.global_code_ch = args.hidden_dim
|
43 |
+
args.netG_use_noise = True
|
44 |
+
args.test_time = (args.test_time == 1)
|
45 |
+
|
46 |
+
if not hasattr(args, 'tex_code_dim'):
|
47 |
+
args.tex_code_dim = 256
|
48 |
+
|
49 |
+
class Tester(TesterBase):
|
50 |
+
def define_model(self):
|
51 |
+
"""Define model
|
52 |
+
"""
|
53 |
+
args = self.args
|
54 |
+
module = importlib.import_module('models.week0417.{}'.format(args.model_name))
|
55 |
+
self.model = module.AE(args)
|
56 |
+
self.model.to(args.device)
|
57 |
+
self.model.eval()
|
58 |
+
return
|
59 |
+
|
60 |
+
def draw_color_seg(self, seg):
|
61 |
+
seg = seg.detach().cpu().numpy()
|
62 |
+
color_ = []
|
63 |
+
for i in range(seg.shape[0]):
|
64 |
+
colori = colorEncode(seg[i].squeeze())
|
65 |
+
colori = torch.from_numpy(colori / 255.0).float().permute(2, 0, 1)
|
66 |
+
color_.append(colori)
|
67 |
+
color_ = torch.stack(color_)
|
68 |
+
return color_
|
69 |
+
|
70 |
+
def to_pil(self, tensor):
|
71 |
+
return transforms.ToPILImage()(tensor.cpu().squeeze().clamp(0.0, 1.0)).convert("RGB")
|
72 |
+
|
73 |
+
def display(self):
|
74 |
+
with st.spinner('Running...'):
|
75 |
+
with torch.no_grad():
|
76 |
+
grouping_mask = self.model_forward(self.data, self.slic, return_type = 'grouping')
|
77 |
+
|
78 |
+
data = (self.data + 1) / 2.0
|
79 |
+
|
80 |
+
seg = grouping_mask.view(-1, 1, args.crop_size, args.crop_size)
|
81 |
+
color_vq = self.draw_color_seg(seg)
|
82 |
+
color_vq = color_vq * 0.8 + data.cpu() * 0.2
|
83 |
+
|
84 |
+
st.markdown('<p class="big-font">Given the image you chose, our model decomposes the image into ten texture segments, each depicts one kind of texture in the image.</p>', unsafe_allow_html=True)
|
85 |
+
col1, col2, col3, col4 = st.columns(4)
|
86 |
+
with col1:
|
87 |
+
st.markdown("")
|
88 |
+
|
89 |
+
with col2:
|
90 |
+
st.markdown("Chosen image")
|
91 |
+
st.image(self.to_pil(data))
|
92 |
+
|
93 |
+
with col3:
|
94 |
+
st.markdown("Grouping mask")
|
95 |
+
st.image(self.to_pil(color_vq))
|
96 |
+
|
97 |
+
with col4:
|
98 |
+
st.markdown("")
|
99 |
+
|
100 |
+
seg_onehot = label2one_hot_torch(seg, C = 10)
|
101 |
+
parts = data.cpu() * seg_onehot.squeeze().unsqueeze(1)
|
102 |
+
|
103 |
+
st.markdown('<p class="big-font">We show all texture segments below. To synthesize an arbitrary-sized texture image from a texture segment, choose and click one of the texture segments below.</p>', unsafe_allow_html=True)
|
104 |
+
tmp_img_list = []
|
105 |
+
for i in range(parts.shape[0]):
|
106 |
+
part_img = self.to_pil(parts[i])
|
107 |
+
out_path = '/home/xli/Dropbox/PAS/tmp/{}.png'.format(i)
|
108 |
+
part_img.save(out_path)
|
109 |
+
|
110 |
+
with open(out_path, "rb") as image:
|
111 |
+
encoded = base64.b64encode(image.read()).decode()
|
112 |
+
tmp_img_list.append(f"data:image/jpeg;base64,{encoded}")
|
113 |
+
|
114 |
+
tex_idx = clickable_images(
|
115 |
+
tmp_img_list,
|
116 |
+
titles=[f"Group #{str(i)}" for i in range(len(tmp_img_list))],
|
117 |
+
div_style={"display": "flex", "justify-content": "center", "flex-wrap": "wrap"},
|
118 |
+
img_style={"margin": "5px", "height": "150px"},
|
119 |
+
key=0
|
120 |
+
)
|
121 |
+
|
122 |
+
if tex_idx > -1:
|
123 |
+
with st.spinner('Running...'):
|
124 |
+
st.markdown('<p class="big-font">You can slide the bar below to set the size of the synthesized texture image.</p>', unsafe_allow_html=True)
|
125 |
+
tex_size = st.slider('', 0, 1000, 256)
|
126 |
+
tex_size = (tex_size // 8) * 8
|
127 |
+
with torch.no_grad():
|
128 |
+
tex = self.model_forward(self.data, self.slic, tex_idx = tex_idx, tex_size = tex_size, return_type = 'tex')
|
129 |
+
col1, col2, col3, col4 = st.columns([1, 1, 4, 1])
|
130 |
+
with col1:
|
131 |
+
st.markdown("")
|
132 |
+
|
133 |
+
with col2:
|
134 |
+
st.markdown("Chosen examplar segment")
|
135 |
+
st.image(self.to_pil(parts[tex_idx]))
|
136 |
+
|
137 |
+
with col3:
|
138 |
+
st.markdown("Synthesized texture image")
|
139 |
+
st.image(self.to_pil(tex))
|
140 |
+
|
141 |
+
with col4:
|
142 |
+
st.markdown("")
|
143 |
+
st.markdown('<p class="big-font">You can choose another image from the examplar images on the top and start again!</p>', unsafe_allow_html=True)
|
144 |
+
#torch.cuda.empty_cache()
|
145 |
+
|
146 |
+
"""
|
147 |
+
st.markdown("#### Texture Editing")
|
148 |
+
st.markdown("**Choose one texture segment to remove.**")
|
149 |
+
remove_idx = clickable_images(
|
150 |
+
tmp_img_list,
|
151 |
+
titles=[f"Group #{str(i)}" for i in range(len(tmp_img_list))],
|
152 |
+
div_style={"display": "flex", "justify-content": "center", "flex-wrap": "wrap"},
|
153 |
+
img_style={"margin": "5px", "height": "120px"},
|
154 |
+
key=1
|
155 |
+
)
|
156 |
+
st.markdown("**Choose one texture segment to fill in the missing pixels.**")
|
157 |
+
fill_idx = clickable_images(
|
158 |
+
tmp_img_list,
|
159 |
+
titles=[f"Group #{str(i)}" for i in range(len(tmp_img_list))],
|
160 |
+
div_style={"display": "flex", "justify-content": "center", "flex-wrap": "wrap"},
|
161 |
+
img_style={"margin": "5px", "height": "120px"},
|
162 |
+
key=2
|
163 |
+
)
|
164 |
+
rec = self.model_forward(self.data, self.slic, return_type = 'editing', fill_idx = fill_idx, remove_idx = remove_idx)
|
165 |
+
st.image(self.to_pil(rec))
|
166 |
+
"""
|
167 |
+
|
168 |
+
def model_forward(self, rgb_img, slic, epoch = 1000, test_time = False,
|
169 |
+
test = True, tex_idx = None, tex_size = 256,
|
170 |
+
return_type = 'tex', fill_idx = None, remove_idx = None):
|
171 |
+
args = self.args
|
172 |
+
B, _, imgH, imgW = rgb_img.shape
|
173 |
+
|
174 |
+
# Encoder: img (B, 3, H, W) -> feature (B, C, imgH//8, imgW//8)
|
175 |
+
conv_feat, _ = self.model.enc(rgb_img)
|
176 |
+
B, C, H, W = conv_feat.shape
|
177 |
+
|
178 |
+
# Texture code for each superpixel
|
179 |
+
tex_code = self.model.ToTexCode(conv_feat)
|
180 |
+
|
181 |
+
code = F.interpolate(tex_code, size = (imgH, imgW), mode = 'bilinear', align_corners = False)
|
182 |
+
pool_code = poolfeat(code, slic, avg = True)
|
183 |
+
|
184 |
+
prop_code, sp_assign, conv_feats = self.model.gcn(pool_code, slic, (args.add_clustering_epoch <= epoch))
|
185 |
+
softmax = F.softmax(sp_assign * args.temperature, dim = 1)
|
186 |
+
if return_type == 'grouping':
|
187 |
+
return torch.argmax(sp_assign.cpu(), dim = 1)
|
188 |
+
|
189 |
+
|
190 |
+
tex_seg = poolfeat(conv_feats, softmax, avg = True)
|
191 |
+
seg = label2one_hot_torch(torch.argmax(softmax, dim = 1).unsqueeze(1), C = softmax.shape[1])
|
192 |
+
|
193 |
+
if return_type == 'tex':
|
194 |
+
sampled_code = tex_seg[:, tex_idx, :]
|
195 |
+
rec_tex = sampled_code.view(1, -1, 1, 1).repeat(1, 1, tex_size, tex_size)
|
196 |
+
sine_wave = self.model.get_sine_wave(rec_tex, 'rec')
|
197 |
+
H = tex_size // 8; W = tex_size // 8
|
198 |
+
noise = torch.randn(B, self.model.sine_wave_dim, H, W).to(tex_code.device)
|
199 |
+
dec_input = torch.cat((sine_wave, noise), dim = 1)
|
200 |
+
|
201 |
+
weight = self.model.ChannelWeight(rec_tex)
|
202 |
+
weight = F.adaptive_avg_pool2d(weight, output_size = (1)).view(weight.shape[0], -1, 1, 1)
|
203 |
+
weight = torch.sigmoid(weight)
|
204 |
+
dec_input *= weight
|
205 |
+
|
206 |
+
rep_rec = self.model.G(dec_input, rec_tex)
|
207 |
+
rep_rec = (rep_rec + 1) / 2.0
|
208 |
+
return rep_rec
|
209 |
+
elif return_type == 'editing':
|
210 |
+
remove_mask = 0
|
211 |
+
fill_mask = 1
|
212 |
+
rec_tex = upfeat(tex_seg, seg)
|
213 |
+
remove_mask = seg[:, remove_idx:remove_idx+1]
|
214 |
+
fill_tex = tex_seg[:, fill_idx, :].view(1, -1, 1, 1).repeat(1, 1, imgH, imgW)
|
215 |
+
rec_tex = rec_tex * (1 - remove_mask) + fill_tex * remove_mask
|
216 |
+
|
217 |
+
sine_wave = self.model.get_sine_wave(rec_tex, 'rec')
|
218 |
+
H = imgH // 8; W = imgW // 8
|
219 |
+
noise = torch.randn(B, self.model.sine_wave_dim, H, W).to(tex_code.device)
|
220 |
+
dec_input = torch.cat((sine_wave, noise), dim = 1)
|
221 |
+
weight = self.model.ChannelWeight(rec_tex)
|
222 |
+
weight = F.adaptive_avg_pool2d(weight, output_size = (1)).view(weight.shape[0], -1, 1, 1)
|
223 |
+
weight = torch.sigmoid(weight)
|
224 |
+
dec_input *= weight
|
225 |
+
|
226 |
+
rep_rec = self.model.G(dec_input, rec_tex)
|
227 |
+
rep_rec = (rep_rec + 1) / 2.0
|
228 |
+
return rep_rec
|
229 |
+
|
230 |
+
|
231 |
+
def load_data(self, data_path):
|
232 |
+
rgb_img = Image.open(data_path)
|
233 |
+
crop_size = self.args.crop_size
|
234 |
+
i = 40; j = 40; h = crop_size; w = crop_size
|
235 |
+
rgb_img = TF.crop(rgb_img, i, j, h, w)
|
236 |
+
|
237 |
+
# compute superpixel
|
238 |
+
sp_num = 196
|
239 |
+
slic_i = slic(np.array(rgb_img), n_segments=sp_num, compactness=10, start_label=0, min_size_factor=0.3)
|
240 |
+
slic_i = torch.from_numpy(slic_i)
|
241 |
+
slic_i[slic_i >= sp_num] = sp_num - 1
|
242 |
+
oh = label2one_hot_torch(slic_i.unsqueeze(0).unsqueeze(0), C = sp_num).squeeze()
|
243 |
+
self.slic = oh.unsqueeze(0).to(args.device)
|
244 |
+
|
245 |
+
rgb_img = TF.to_tensor(rgb_img)
|
246 |
+
rgb_img = rgb_img.unsqueeze(0)
|
247 |
+
self.data = rgb_img.to(args.device) * 2 - 1
|
248 |
+
|
249 |
+
def load_model(self, model_path):
|
250 |
+
self.model = torch.nn.DataParallel(self.model)
|
251 |
+
cpk = torch.load(model_path)
|
252 |
+
saved_state_dict = cpk['model']
|
253 |
+
self.model.load_state_dict(saved_state_dict)
|
254 |
+
self.model = self.model.module
|
255 |
+
return
|
256 |
+
|
257 |
+
def test(self):
|
258 |
+
""" Test function
|
259 |
+
"""
|
260 |
+
#for iteration in tqdm(range(args.nsamples)):
|
261 |
+
self.test_step(0)
|
262 |
+
self.display(0, 'train')
|
263 |
+
|
264 |
+
def main():
|
265 |
+
#torch.cuda.empty_cache()
|
266 |
+
st.set_page_config(layout="wide")
|
267 |
+
st.markdown("""
|
268 |
+
<style>
|
269 |
+
.big-font {
|
270 |
+
font-size:30px !important;
|
271 |
+
}
|
272 |
+
</style>
|
273 |
+
""", unsafe_allow_html=True)
|
274 |
+
|
275 |
+
st.title("Scraping Textures from Natural Images for Synthesis and Editing")
|
276 |
+
#st.markdown("**In this demo, we show how to scrape textures from natural images for texture synthesis and editing.**")
|
277 |
+
st.markdown('<p class="big-font">In this demo, we show how to scrape textures from natural images for texture synthesis and editing.</p>', unsafe_allow_html=True)
|
278 |
+
st.markdown("## Texture synthesis")
|
279 |
+
st.markdown('<p class="big-font">Here we provide a set of example images, please choose and click one image to start.</p>', unsafe_allow_html=True)
|
280 |
+
img_list = glob(os.path.join("data/images/*.jpg"))
|
281 |
+
test_img_list = glob(os.path.join("data/test_images/*.jpg"))
|
282 |
+
img_list.extend(test_img_list)
|
283 |
+
byte_img_list = []
|
284 |
+
for img_path in img_list:
|
285 |
+
with open(img_path, "rb") as image:
|
286 |
+
encoded = base64.b64encode(image.read()).decode()
|
287 |
+
byte_img_list.append(f"data:image/jpeg;base64,{encoded}")
|
288 |
+
img_idx = clickable_images(
|
289 |
+
byte_img_list,
|
290 |
+
titles=[f"Group #{str(i)}" for i in range(len(byte_img_list))],
|
291 |
+
div_style={"display": "flex", "justify-content": "center", "flex-wrap": "wrap"},
|
292 |
+
img_style={"margin": "5px", "height": "150px"},
|
293 |
+
)
|
294 |
+
img_path = img_list[img_idx]
|
295 |
+
|
296 |
+
img_name = img_path.split("/")[-1]
|
297 |
+
args.pretrained_path = os.path.join("/home/xli/WORKDIR/04-18/{}/cpk.pth".format(img_name.split(".")[0]))
|
298 |
+
|
299 |
+
if img_idx > -1:
|
300 |
+
tester = Tester(args)
|
301 |
+
tester.define_model()
|
302 |
+
tester.load_data(img_path)
|
303 |
+
tester.load_model(args.pretrained_path)
|
304 |
+
tester.display()
|
305 |
+
|
306 |
+
if __name__ == '__main__':
|
307 |
+
main()
|