-
Notifications
You must be signed in to change notification settings - Fork 1
/
DreamPhant.py
164 lines (136 loc) · 7.35 KB
/
DreamPhant.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
_author__ = 'MSteger'
import numpy as np
import torch
import PIL
import os, gc
import scipy.ndimage as nd
import PIL.Image
from torch.autograd import Variable
from torchvision import transforms, models
from tqdm import tqdm
class DreamPhant(object):
def __init__(self, model, input_dir, device = torch.device('cpu'), step_fn=None, verbose = True):
self.model = model.to(device)
self.input_dir = input_dir
self.device = device
self.step_fn = self.make_step if step_fn is None else step_fn
self.verbose = verbose
def _load_image(self, path, preprocess, resize = None):
img = PIL.Image.open(path)
if resize is not None: img.thumbnail(resize, PIL.Image.ANTIALIAS)
img_tensor = preprocess(img).unsqueeze(0) if preprocess is not None else transforms.ToTensor(img)
return img, img_tensor, img_tensor.numpy()
def _data_to_img(self, t, tensor = True):
if tensor: t = t.numpy()
mean = np.array([0.485, 0.456, 0.406]).reshape([1, 1, 3])
std = np.array([0.229, 0.224, 0.225]).reshape([1, 1, 3])
inp = t[0, :, :, :]
inp = inp.transpose(1, 2, 0)
inp = std * inp + mean
inp *= 255
inp = np.uint8(np.clip(inp, 0, 255))
return PIL.Image.fromarray(inp)
def _image_to_variable(self, image, requires_grad=False):
return Variable(image.cuda() if self.device == torch.device('cuda') else image, requires_grad=requires_grad)
def _extract_features(self, img_tensor, layer, model = None):
if model is None: model = self.model
features = self._image_to_variable(img_tensor, requires_grad=True) if not isinstance(img_tensor, (torch.cuda.FloatTensor if self.device == torch.device('cuda') else torch.Tensor)) else img_tensor
for index, current_layer in enumerate(model.features.children()):
features = current_layer(features)
if index == layer: break
return features
def objective(self, dst, guide_features=None):
if guide_features is None:
return dst.data
else:
x = dst.data[0].cpu().numpy()
y = guide_features.data[0].cpu().numpy()
ch, w, h = x.shape
x = x.reshape(ch, -1)
y = y.reshape(ch, -1)
A = x.T.dot(y)
diff = y[:, A.argmax(1)]
return torch.Tensor(np.array([diff.reshape(ch, w, h)])).to(self.device)
def make_step(self, img, control=None, step_size=1.5, layer=28, jitter=32):
mean = np.array([0.485, 0.456, 0.406]).reshape([3, 1, 1])
std = np.array([0.229, 0.224, 0.225]).reshape([3, 1, 1])
ox, oy = np.random.randint(-jitter, jitter + 1, 2) # offset by random jitter
img = np.roll(np.roll(img, ox, -1), oy, -2) # apply jitter shift
tensor = torch.Tensor(img)
img_var = self._image_to_variable(tensor, requires_grad=True)
self.model.zero_grad()
x = self._extract_features(img_tensor=img_var, layer=layer)
delta = self.objective(x, control)
x.backward(delta)
# L2 Regularization on gradients
mean_square = torch.Tensor([torch.mean(img_var.grad.data ** 2)]).to(self.device)
img_var.grad.data /= torch.sqrt(mean_square)
img_var.data.add_(img_var.grad.data * step_size)
result = img_var.data.cpu().numpy()
result = np.roll(np.roll(result, -ox, -1), -oy, -2)
result[0, :, :, :] = np.clip(result[0, :, :, :], -mean / std, (1 - mean) / std) # normalize img
return torch.Tensor(result)
def DeepDream(self, base_img, octave_n=6, octave_scale=1.4, iter_n=10, **step_args):
octaves = [base_img]
for i in range(octave_n - 1): octaves.append(nd.zoom(octaves[-1], (1, 1, 1.0 / octave_scale, 1.0 / octave_scale), order=1))
detail = np.zeros_like(octaves[-1])
for octave, octave_base in enumerate(octaves[::-1]):
h, w = octave_base.shape[-2:]
if octave > 0:
h1, w1 = detail.shape[-2:]
detail = nd.zoom(detail, (1, 1, 1.0 * h / h1, 1.0 * w / w1), order=1)
src = octave_base + detail
for i in range(iter_n):
src = self.step_fn(src, **step_args)
detail = src.numpy() - octave_base
return src
def transform(self, preprocess, layer, control=None, resize = [1024, 1024], repeated = 10, file_prefix = None,**dream_args):
if (repeated is None) | (repeated is False): repeated = 1
if control is not None:
_, guideImage_tensor, _ = self._load_image(path=control[2], preprocess=control[3], resize=control[4])
control = self._extract_features(img_tensor=guideImage_tensor, layer = control[0])
for img_name in os.listdir(self.input_dir):
img_path = os.path.join(self.input_dir, img_name)
_, _, frame = self._load_image(path=img_path, preprocess=preprocess, resize=resize)
bar = tqdm(total=repeated, unit='iteration')
for i in range(repeated):
frame = self.DeepDream(base_img=frame, layer = layer, control = control, **dream_args).numpy()
bar.update(1)
bar.set_description('Processing Image: {}'.format(img_name))
bar.close()
DeepDream = self._data_to_img(frame, tensor=False)
output_dir = os.path.join(self.input_dir.replace('/input', '/output'), 'layer{}'.format(layer))
if not os.path.exists(output_dir): os.makedirs(output_dir)
if file_prefix is not None: img_name = '{}_{}'.format(file_prefix, img_name)
output_path = os.path.join(output_dir, img_name)
DeepDream.save(output_path)
if self.verbose: print 'saved img {} to {}'.format(os.path.split(img_name)[-1], output_path)
return self
if __name__ == '__main__':
from utils.helpers import summary
print torch.__version__
# setup
input_dir = r'/media/msteger/storage/resources/DreamPhant/dream/input/'
guideImage_dir = r'/media/msteger/storage/resources/DreamPhant/dream/guides/selected/'
preprocess = transforms.Compose([transforms.ToTensor(), transforms.Normalize(mean=[0.485, 0.456, 0.406],std=[0.229, 0.224, 0.225])])
preprocess_resize = transforms.Compose([transforms.Resize((224, 224)),transforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
device = torch.device('cuda')
iter_n = 5
step_size = 0.01
jitter = 32
guided = True
# model
model = models.vgg19(pretrained=True)
for weights in model.parameters(): weights.requires_grad = False
summary(model=model, device=device, input_size=(1,3,224,224))
# dreaming
for guideImage_name in os.listdir(guideImage_dir):
guideImage_path = os.path.join(guideImage_dir, guideImage_name)
guideImage_name = guideImage_name.split('.jpg')[0]
for rep in range(30, 120, 30):
Dream = DreamPhant(model=model, input_dir=input_dir, device=device)
Dream.transform(preprocess = preprocess, resize = [768, 1024], layer = 20, octave_n=6, octave_scale=1.4,iter_n=iter_n, control=(model, 20, guideImage_path, preprocess_resize, None) if guided else None,\
step_size=step_size, jitter=jitter, repeated = rep, file_prefix='{}_{}_{}_{}_{}_{}'.format(guideImage_name, rep, iter_n, step_size, jitter, guided))
Dream = None
gc.collect()
print 'done'