/
inception_score.py
188 lines (169 loc) · 5.93 KB
/
inception_score.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
import math
import torch
from torch import nn
import torch.nn.functional as F
from torchvision import transforms
from torchvision.models import inception_v3
import numpy as np
def get_inception_score(imgs, use_cuda=None):
add_channels = transforms.Compose([
transforms.ToPILImage(),
transforms.Grayscale(3),
transforms.ToTensor()])
up = nn.Upsample(size=(299, 299), mode='bilinear')
net = inception_v3(pretrained=True)
# net.eval()
if torch.cuda.is_available() and use_cuda != False:
net = net.cuda()
elif (torch.cuda.is_available() == False) and use_cuda == False:
print("not using cuda")
use_cuda = False
else:
print("Cuda not availiabe but use_cuda is True")
return
batch_size = np.shape(imgs[0])[0]
assert(len(np.shape(imgs[0])) == 4), "Batches of imgs had incorrect number of dimensions. Expected 5. Recieved shape: " + str(np.shape(imgs))
scores = []
for batch in imgs:
batch_with_channels = torch.zeros((batch_size,3, 32, 32))
for i in range(len(batch)):
img = batch[i,:,:]
curr_img = img.detach().cpu().numpy().T
img = add_channels(curr_img).squeeze()
batch_with_channels[i,:,:] = img
batch = batch_with_channels
batch = up(batch)
s,_ = net(batch)
scores += [s]
print("scores calculated")
p_yx = F.softmax(torch.cat(scores, 0), 1)
p_y = p_yx.mean(0).unsqueeze(0).expand(p_yx.size(0), -1)
KL_d = p_yx * (torch.log(p_yx) - torch.log(p_y))
final_score = KL_d.mean()
final_score = float(final_score.detach().cpu().numpy())
print("inception score", final_score)
return final_score
# import math
# import torch
# from torch import nn
# import torch.nn.functional as F
# from torchvision import transforms
# from torchvision.models import inception_v3
# import numpy as np
# def get_inception_score(imgs, use_cuda=None):
# add_channels = transforms.Compose([
# transforms.ToPILImage(),
# transforms.Grayscale(3),
# transforms.ToTensor()])
# up = nn.Upsample(size=(299, 299), mode='bilinear')
# net = inception_v3(pretrained=True)
# net.eval()
# if torch.cuda.is_available() and use_cuda != False:
# net = net.cuda()
# elif (torch.cuda.is_available() == False) and use_cuda == False:
# use_cuda = False
# else:
# print("Cuda not availiabe but use_cuda is True")
# return
# batch_size = np.shape(imgs[0])[0]
# print(batch_size)
# assert(len(np.shape(imgs[0])) == 4), "Batches of imgs had incorrect number of dimensions. Expected 5. Recieved shape: " + str(np.shape(imgs))
# # scores = []
# n = batch_size * len(imgs)
# scores = np.zeros((n,1000))
# print(scores.shape)
# scores = torch.zeros((n,1000))
# for batch in imgs:
# batch_with_channels = torch.zeros((batch_size,3, 32, 32))
# for i in range(batch_size):
# # img = batch[i,:,:]
# curr_img = batch[i]
# curr_img = curr_img.detach().cpu().numpy().T
# curr_img = add_channels(curr_img).squeeze()
# # batch_with_channels[i,:,:] = img
# batch_with_channels[i] = curr_img
# batch = batch_with_channels
# batch = up(batch)
# s = net(batch)
# print(s.shape)
# # s = s.data.cpu().numpy()
# # s = s.reshape((-1, 1000))
# scores[i*batch_size: (i*batch_size) + batch_size] = s
# print("scores", scores)
# p_yx = F.softmax(torch.cat(scores, 0), 1)
# p_y = p_yx.mean(0).unsqueeze(0).expand(p_yx.size(0), -1)
# KL_d = p_yx * (torch.log(p_yx) - torch.log(p_y))
# final_score = KL_d.mean()
# final_score = float(final_score.detach().cpu().numpy())
# print("inception score", final_score)
# return final_score
############# Another implementation
# import torch
# from torch import nn
# from torch.autograd import Variable
# from torch.nn import functional as F
# import torch.utils.data
# from torchvision.models.inception import inception_v3
# import numpy as np
# from scipy.stats import entropy
# def get_inception_score(imgs, cuda=True, resize=True, splits=1):
# """Computes the inception score of the generated images imgs
# # imgs -- Torch dataset of (3xHxW) numpy images normalized in the range [-1, 1]
# cuda -- whether or not to run on GPU
# batch_size -- batch size for feeding into Inception v3
# splits -- number of splits
# """
# # print(len(imgs))
# # print(imgs[])
# batch_size = ((imgs[0]).shape)[0]
# N = len(imgs) * batch_size
# # Set up dtype
# if cuda:
# dtype = torch.cuda.FloatTensor
# else:
# if torch.cuda.is_available():
# print("WARNING: You have a CUDA device, so you should probably set cuda=True")
# dtype = torch.FloatTensor
# # Set up dataloader
# # dataloader = torch.utils.data.DataLoader(imgs, batch_size=batch_size)
# # Load inception model
# inception_model = inception_v3(pretrained=True, transform_input=False).type(dtype)
# inception_model.eval();
# up = nn.Upsample(size=(299, 299), mode='bilinear').type(dtype)
# add_channels = transforms.Compose([
# transforms.ToPILImage(),
# transforms.Grayscale(3),
# transforms.ToTensor()])
# def get_pred(x):
# x = up(x)
# x = inception_model(x)
# return F.softmax(x).data.cpu().numpy()
# # Get predictions
# preds = np.zeros((N, 1000))
# # for i, batch in enumerate(dataloader, 0):
# for i, batch in enumerate(imgs):
# batch_with_channels = torch.zeros((batch_size,3, 32, 32))
# for i in range(batch_size):
# curr_img = batch[i]
# curr_img = curr_img.detach().cpu().numpy().T
# curr_img = add_channels(curr_img).squeeze()
# batch_with_channels[i] = curr_img
# batch = batch_with_channels
# batch = batch.type(dtype)
# batchv = Variable(batch)
# batch_size_i = batch.size()[0]
# preds[i*batch_size:i*batch_size + batch_size_i] = get_pred(batchv)
# # Now compute the mean kl-div
# split_scores = []
# for k in range(splits):
# part = preds[k * (N // splits): (k+1) * (N // splits), :]
# py = np.mean(part, axis=0)
# scores = []
# for i in range(part.shape[0]):
# pyx = part[i, :]
# print(pyx) #all zeros
# # print(py)
# scores.append(entropy(pyx, py))
# split_scores.append(np.exp(np.mean(scores)))
# print("mean score", np.mean(split_scores))
# return np.mean(split_scores) #, np.std(split_scores)