-
Notifications
You must be signed in to change notification settings - Fork 5
/
utils.py
151 lines (113 loc) · 4.98 KB
/
utils.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
import os
import pandas as pd
import torch
import numpy as np
from torch.utils.data import Dataset
from image_lib import resize_to_square
import cv2
from PIL import Image
import matplotlib.pyplot as plt
import torch.nn.functional as F
import config
from sklearn.preprocessing import MinMaxScaler
normalization_values = np.load(config.NORMALIZATION_VALUES_PATH)
class CustomImageDataset(Dataset):
def __init__(self, annotations_file, img_dir, transform=None, target_transform=None, use_saved_feat=False):
self.json_df = pd.read_csv(annotations_file)
self.img_dir = img_dir
self.use_saved_feat = use_saved_feat
categ_dict = torch.load(config.CATEG_DICT)
color_dict = torch.load(config.COLOR_DICT)
fabric_dict = torch.load(config.FAB_DICT)
# Category
self.category = [categ_dict[x] for x in self.json_df['category'].tolist()]
# Color
self.color = [color_dict[x] for x in self.json_df['exact_color'].tolist()]
# Fabric
self.fabric = [fabric_dict[x] for x in self.json_df['texture'].tolist()]
# Release date
self.release_date = self.json_df['release_date'].tolist()
#Days
self.days = self.json_df['day'].tolist()
#Weeks
self.weeks = self.json_df['week'].tolist()
#Months
self.months = self.json_df['month'].tolist()
#Years
self.years = self.json_df['year'].tolist()
# Labels
new_labels = self.json_df.iloc[:, 1:13].values.tolist()
self.img_labels = pd.Series(new_labels)
# Path
self.path = self.json_df["image_path"]
# Transform
self.transform = transform
self.target_transform = target_transform
def __len__(self):
return len(self.img_labels)
def __getitem__(self, idx):
# Image
img_path = os.path.join(self.img_dir, self.path.iloc[idx])
image = cv2.imread(img_path)
image = resize_to_square(image)
image_2 = cv2.cvtColor(image, cv2.COLOR_RGB2RGBA)
orig_8x8 = cv2.resize(image_2, (8,8), interpolation = cv2.INTER_AREA)
# Image feature
feat_path = os.path.join(config.SAVED_FEATURES_PATH, self.path.iloc[idx].replace(".png", ".pth"))
img_feature=torch.load(feat_path).squeeze()
# Category
category = self.category[idx]
# Color
color = self.color[idx]
# Fabric
fabric = self.fabric[idx]
# Release date
release_date = self.release_date[idx]
#Temporal features
temporal_features = []
temporal_features.append(self.days[idx])
temporal_features.append(self.weeks[idx])
temporal_features.append(self.months[idx])
temporal_features.append(self.years[idx])
temporal_features = torch.as_tensor(temporal_features, dtype=torch.float)
# Label
trend = self.img_labels.iloc[idx]
trend = torch.FloatTensor(trend)
# Applying transform
if self.transform:
image_transformed = self.transform(image)
if self.target_transform:
trend = self.target_transform(trend)
return (image_transformed, trend, category, color, fabric, orig_8x8, release_date, temporal_features, img_feature, self.path.iloc[idx])
def resize2d(img, size):
from torch.autograd import Variable
return (F.adaptive_avg_pool2d(Variable(img,volatile=True), size)).data
def exog_extractor(date, categ, color, fabric):
categ = np.asarray(categ)
color = np.asarray(color)
fabric = np.asarray(fabric)
gtrends = pd.read_csv(config.COMPOSED_GTREND, parse_dates=['date'], index_col=[0])
out_gtrends = []
weeks = config.EXOG_LEN
for i in range(categ.shape[0]):
categ_dict = torch.load(config.CATEG_DICT)
categ_dict = {v: k for k, v in categ_dict.items()}
color_dict = torch.load(config.COLOR_DICT)
color_dict = {v: k for k, v in color_dict.items()}
fabric_dict = torch.load(config.FAB_DICT)
fabric_dict = {v: k for k, v in fabric_dict.items()}
cat = categ_dict[categ[i]]
col = color_dict[color[i]]
fab = fabric_dict[fabric[i]]
start_date = pd.to_datetime(date[i])
gtrend_start = start_date - pd.DateOffset(weeks=52)
cat_gtrend = gtrends.loc[gtrend_start:start_date][cat][-52:].values
col_gtrend = gtrends.loc[gtrend_start:start_date][col][-52:].values
fab_gtrend = gtrends.loc[gtrend_start:start_date][fab.replace(' ', '')][-52:].values
cat_gtrend = MinMaxScaler().fit_transform(cat_gtrend.reshape(-1,1)).flatten()
col_gtrend = MinMaxScaler().fit_transform(col_gtrend.reshape(-1,1)).flatten()
fab_gtrend = MinMaxScaler().fit_transform(fab_gtrend.reshape(-1,1)).flatten()
multitrends = np.hstack([cat_gtrend[:weeks], col_gtrend[:weeks], fab_gtrend[:weeks]]).astype(np.float32)
out_gtrends.append(multitrends)
out_gtrends = np.vstack(out_gtrends)
return out_gtrends