/
attention_map.py
252 lines (214 loc) · 9.03 KB
/
attention_map.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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
import os
import torch
import torch.optim as optim
import torch.nn as nn
from core.DAZLE_plot import DAZLE
from core.CUBDataLoader import CUBDataLoader
from core.helper_func import eval_zs_gzsl
from global_setting import NFS_path
import numpy as np
import wandb
from get_gpu_info import get_gpu_info
from PIL import Image
import matplotlib.pyplot as plt
import skimage
from sklearn.manifold import TSNE
from torchvision import transforms
data_transforms = transforms.Compose([
transforms.Resize(448),
transforms.CenterCrop(448),
transforms.ToTensor()])
def dazle_visualize_attention_np_global_448(img_ids,alphas_1,alphas_2,attr_name,save_path=None):
# alphas_1: [bir] alphas_2: [bi]
n = img_ids.shape[0]
image_size = 448 #one side of the img
assert alphas_1.shape[1] == alphas_2.shape[1] == len(attr_name)
r = alphas_1.shape[2]
h = w = int(np.sqrt(r))
for i in range(n):
fig=plt.figure(i,figsize=(33, 5))
file_path=img_ids[i]#.decode('utf-8')
img_name = file_path.split("/")[-1]
alpha_1 = alphas_1[i] #[ir]
alpha_2 = alphas_2[i] #[i]
# score = S[i]
# Plot original image
image = Image.open(file_path)
if image.mode == 'L':
image=image.convert('RGB')
image = data_transforms(image)
image = image.permute(1,2,0) #[224,244,3] <== [3,224,224]
idx = 1
ax = plt.subplot(1, 11, 1)
idx += 1
plt.imshow(image)
# ax.set_title(os.path.splitext(img_name)[0],{'fontsize': 13})
plt.axis('off')
idxs_top_p=np.argsort(-alpha_2)[:10]
idxs_top_g=np.argsort(-alpha_2)[:200]
# idxs_top_n=np.argsort(alpha_2)[:3]
#pdb.set_trace()
for idx_ctxt,idx_attr in enumerate(idxs_top_p):
ax=plt.subplot(1, 11, idx)
idx += 1
plt.imshow(image)
alp_curr = alpha_1[idx_attr,:].reshape(14,14)
alp_img = skimage.transform.pyramid_expand(alp_curr, upscale=image_size/h, sigma=10,multichannel=False)
plt.imshow(alp_img, alpha=0.5, cmap='jet')
# ax.set_title("{}\n{}\n{}-{}".format(attr_name[idx_attr],alpha_2[idx_attr],score[idx_attr],attr[idx_attr]),{'fontsize': 10})
# ax.set_title("{}\n(Score = {:.2f})".format(attr_name[idx_attr].title().replace(
# ' ', ''), alpha_2[idx_attr]), {'fontsize': 19})
ax.set_title("{}\n(Score = {:.1f})".format(' '.join(attr_name[idx_attr].split()[:2]).title(
) + '\n' + ' '.join(attr_name[idx_attr].split()[2:]).title(), alpha_2[idx_attr]), {'fontsize': 25})
plt.axis('off')
fig.tight_layout()
if save_path is not None:
plt.savefig(save_path+img_name,dpi=200)
plt.close()
def dazle_visualize_attention_np_global_448_small(img_ids,alphas_1,alphas_2,attr_name,save_path=None):
# alphas_1: [bir] alphas_2: [bi]
n = img_ids.shape[0]
image_size = 448 #one side of the img
assert alphas_1.shape[1] == alphas_2.shape[1] == len(attr_name)
r = alphas_1.shape[2]
h = w = int(np.sqrt(r))
for i in range(n):
fig=plt.figure(i,figsize=(33, 4))
file_path=img_ids[i]#.decode('utf-8')
img_name = file_path.split("/")[-1]
alpha_1 = alphas_1[i] #[ir]
alpha_2 = alphas_2[i] #[i]
# score = S[i]
# Plot original image
image = Image.open(file_path)
if image.mode == 'L':
image=image.convert('RGB')
image = data_transforms(image)
image = image.permute(1,2,0) #[224,244,3] <== [3,224,224]
idx = 1
ax = plt.subplot(1, 11, 1)
idx += 1
plt.imshow(image)
# ax.set_title(os.path.splitext(img_name)[0],{'fontsize': 13})
plt.axis('off')
idxs_top_p=np.argsort(-alpha_2)[:10]
idxs_top_g=np.argsort(-alpha_2)[:200]
# idxs_top_n=np.argsort(alpha_2)[:3]
#pdb.set_trace()
for idx_ctxt,idx_attr in enumerate(idxs_top_p):
ax=plt.subplot(1, 11, idx)
idx += 1
plt.imshow(image)
alp_curr = alpha_1[idx_attr,:].reshape(14,14)
alp_img = skimage.transform.pyramid_expand(alp_curr, upscale=image_size/h, sigma=10,multichannel=False)
plt.imshow(alp_img, alpha=0.5, cmap='jet')
# ax.set_title("{}\n{}\n{}-{}".format(attr_name[idx_attr],alpha_2[idx_attr],score[idx_attr],attr[idx_attr]),{'fontsize': 10})
ax.set_title("{}\n(Score = {:.2f})".format(attr_name[idx_attr].title().replace(
' ', ''), alpha_2[idx_attr]), {'fontsize': 18})
# ax.set_title("{}\n(Score = {:.1f})".format(' '.join(attr_name[idx_attr].split()[:2]).title(
# ) + '\n' + ' '.join(attr_name[idx_attr].split()[2:]).title(), alpha_2[idx_attr]), {'fontsize': 20})
plt.axis('off')
fig.tight_layout()
if save_path is not None:
plt.savefig(save_path+img_name,dpi=200)
plt.close()
def plot_att(config):
model_path = 'saved_model/CUB_weights_H-0.688.pth'
config.dataset = 'CUB'
config.num_class = 200
config.num_attribute = 312
if config.img_size == 224: config.resnet_region = 49
elif config.img_size == 448: config.resnet_region = 196
print('Config file from wandb:', config)
if config.device == 'auto':
device = get_gpu_info()
else:
device = config.device
dataloader = CUBDataLoader(NFS_path, device,
is_unsupervised_attr=False, is_balance=False,
img_size=config.img_size, use_unzip=config.use_unzip)
dataloader.augment_img_path()
torch.backends.cudnn.benchmark = True
def get_lr(optimizer):
lr = []
for param_group in optimizer.param_groups:
lr.append(param_group['lr'])
return lr
seed = config.random_seed
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
batch_size = config.batch_size
nepoches = config.epochs
niters = dataloader.ntrain * nepoches//batch_size
dim_f = 2048
dim_v = 300
init_w2v_att = dataloader.w2v_att
att = dataloader.att
normalize_att = dataloader.normalize_att
trainable_w2v = config.trainable_w2v
# CE loss和cal loss的超参数
lambda_ = config.lambda_
bias = 0
prob_prune = 0
# uniform DAZLE attention的选项
uniform_att_1 = False
uniform_att_2 = False
seenclass = dataloader.seenclasses
unseenclass = dataloader.unseenclasses
desired_mass = 1
report_interval = niters//nepoches
model = DAZLE(config, dim_f,dim_v,init_w2v_att,att,normalize_att,
seenclass,unseenclass,
lambda_,
trainable_w2v,normalize_V=False,normalize_F=True,is_conservative=True,
uniform_att_1=uniform_att_1,uniform_att_2=uniform_att_2,
prob_prune=prob_prune,desired_mass=desired_mass, is_conv=False,
is_bias=config.is_bias)
model.load_state_dict(torch.load(model_path))
model.to(device)
num_parameters = sum([p.numel() for p in model.parameters()]) * 1e-6
print('model parameters: %.3fM' % num_parameters)
file_list = [
'Acadian_Flycatcher_0008_795599',
'American_Goldfinch_0092_32910',
'Canada_Warbler_0117_162394',
'Carolina_Wren_0006_186742',
'Vesper_Sparrow_0090_125690',
'Western_Gull_0058_53882',
'White_Throated_Sparrow_0128_128956',
'Winter_Wren_0118_189805',
'Yellow_Breasted_Chat_0044_22106',
'Elegant_Tern_0085_151091',
'European_Goldfinch_0025_794647',
'Florida_Jay_0008_64482',
'Fox_Sparrow_0025_114555',
'Grasshopper_Sparrow_0053_115991',
'Grasshopper_Sparrow_0107_116286',
'Gray_Crowned_Rosy_Finch_0036_797287'
]
for filename in file_list:
for i, id in enumerate(dataloader.seenclasses):
# if i == 5:
# raise Exception
id = id.item()
(batch_label, batch_feature, batch_files, batch_att) = dataloader.next_batch_img(
batch_size=10, class_id=id, is_trainset=False)
if filename not in str(batch_files):
continue
idx = [filename in str(f) for f in batch_files]
batch_feature = batch_feature[idx]
batch_files = batch_files[idx]
model.eval()
with torch.no_grad():
out_package = model(batch_feature)
# attention map of DAZLE
dazle_visualize_attention_np_global_448_small(batch_files,
out_package['att'].cpu().numpy(),
out_package['dazle_embed'].cpu().numpy(),
dataloader.attr_name,
'plot/atten_fig/')
if __name__ == '__main__':
wandb.init(project='ZSL_DALZE_Transformer_GA', config='config_cub.yaml', allow_val_change=True)
config = wandb.config
plot_att(config)