-
Notifications
You must be signed in to change notification settings - Fork 122
/
deepfashion_datasets.py
221 lines (180 loc) · 8.17 KB
/
deepfashion_datasets.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
import torch.utils.data as data
from PIL import Image
import numpy as np
import cv2
import torchvision.transforms as transforms
import torch
import copy, os, collections
import json
from .human_parse_labels import get_label_map, DF_LABEL, YF_LABEL
import pandas as pd
from utils import pose_utils
TEST_PATCHES = [
'chequered/chequered_0052.jpg','dotted/dotted_0072.jpg',
'paisley/paisley_0015.jpg','striped/striped_0011.jpg'
]
class DFPairDataset(data.Dataset):
def get_paths(self, root, phase, viton=False):
pairLst = os.path.join(root, 'fasion-pairs-%s.csv' % phase)
if viton:
pairLst = os.path.join(root, 'fasion-pairs-%s.csv' % 'viton')
name_pairs = self.init_categories(pairLst)
image_dir = os.path.join(root, '%s' % phase)
bonesLst = os.path.join(root, 'fasion-annotation-%s.csv' % phase)
return image_dir, bonesLst, name_pairs
def init_categories(self, pairLst):
with open(pairLst) as f:
anns = f.readlines()
anns = [line[:-1].split(",")[1:] for line in anns[1:]]
return anns
def __init__(self, dataroot, dim=(256,256), isTrain=True, n_human_part=8, viton=False):
super(DFPairDataset, self).__init__()
self.root = dataroot
self.isTrain = isTrain
self.split = 'train' if isTrain else 'test'
self.n_human_part = n_human_part
self.dim = dim
self._init(viton)
self.mask_dir = self.root + "/%sM_lip" % ('train' if isTrain else 'test')
def _init(self, viton):
self.image_dir, self.bone_file, self.name_pairs = self.get_paths(self.root, self.split, viton)
self.annotation_file = pd.read_csv(self.bone_file, sep=':')
self.annotation_file = self.annotation_file.set_index('name')
self.aiyu2atr, self.atr2aiyu = get_label_map(self.n_human_part)
self.load_size = self.dim
self.crop_size = self.load_size
# transforms
self.resize = transforms.Resize(self.crop_size)
self.toTensor = transforms.ToTensor()
self.toPIL = transforms.ToPILImage()
self.normalize = transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))
def __len__(self):
return len(self.name_pairs)
def _load_img(self, fn):
img = Image.open(fn).convert("RGB")
img = self.resize(img)
img = self.toTensor(img)
img = self.normalize(img)
return img
def _load_mask(self, fn):
mask = Image.open(fn + ".png")
mask = self.resize(mask)
mask = torch.from_numpy(np.array(mask))
texture_mask = copy.deepcopy(mask)
for atr in self.atr2aiyu:
aiyu = self.atr2aiyu[atr]
texture_mask[mask == atr] = aiyu
return texture_mask
def _load_kpt(self, name):
string = self.annotation_file.loc[name]
array = pose_utils.load_pose_cords_from_strings(string['keypoints_y'], string['keypoints_x'])
pose = pose_utils.cords_to_map(array, self.load_size, (256, 176))
pose = np.transpose(pose,(2, 0, 1))
pose = torch.Tensor(pose)
return pose
def get_to_item(self, key):
img = self._load_img(os.path.join(self.image_dir,key))
kpt = self._load_kpt(key)
parse = self._load_mask(os.path.join(self.mask_dir, key[:-4]))
return img, kpt, parse
def __getitem__(self, index):
from_key, to_key = self.name_pairs[index]
from_img, from_kpt, from_parse = self.get_to_item(from_key)
to_img, to_kpt, to_parse = self.get_to_item(to_key)
return from_img, from_kpt, from_parse, to_img, to_kpt, to_parse, index #torch.Tensor([0])
class DFVisualDataset(DFPairDataset):
def __init__(self, dataroot, dim=(256,256), texture_dir="",isTrain=False, n_human_part=8):
DFPairDataset.__init__(self, dataroot, dim, isTrain, n_human_part=n_human_part)
# load anns
# import pdb; pdb.set_trace()
eval_anns_path = os.path.join(dataroot,"standard_test_anns.txt")
self._load_visual_anns(eval_anns_path)
# load standard pose
self._load_standard_pose()
#load standard patches
#patch_root = "/".join(dataroot.split("/")[:-1])
#self.standard_patches = [self._load_img(os.path.join(patch_root, "dtd/images", fn)).unsqueeze(0) for fn in TEST_PATCHES]
#self.standard_patches = torch.cat(self.standard_patches)
self.selected_keys = [ "gfla", "jacket", "lace", "pattern", "plaid", "plain", "print", "strip", "flower"]
self.image_dir = dataroot + "/test"
self.mask_dir = dataroot + "/testM_lip"
def _load_standard_pose(self):
self.standard_poses = [
self._load_kpt(key).unsqueeze(0) for key in self.pose_keys
]
self.standard_poses = torch.cat(self.standard_poses)
def get_patches(self):
return self.standard_patches
def __len__(self):
return sum([len(self.attr_keys[i]) for i in self.attr_keys])
def _load_visual_anns(self, eval_anns_path):
with open(eval_anns_path) as f:
raw_anns = f.readlines()
pose_cnt = 1
self.pose_keys = []
for line in raw_anns[1:]:
if line.startswith("attr"):
break
self.pose_keys.append(line[:-1])
pose_cnt += 1
self.attr_keys = collections.defaultdict(list)
#import pdb; pdb.set_trace()
for line in raw_anns[pose_cnt+1:]:
category, key = line[:-1].split(", ")
self.attr_keys[category].append(key)
mixed = []
for category in ['flower','plaid','print','strip']:
mixed.append(self.attr_keys[category][0])
self.attr_keys["mixed"] = mixed
def get_patch_input(self):
return torch.cat(self.standard_patches)
def get_all_pose(self, key, std_pose=True):
if std_pose:
return self.standard_poses
folder_path = os.path.join(self.kpt_dir,key).split("/")
prefix = folder_path[-1]
folder_path = "/".join(folder_path[:-1])
ret = []
for fn in os.listdir(folder_path):
if fn.startswith(prefix) and fn.endswith('_kpt.npy'):
curr = self._load_kpt(os.path.join(folder_path, fn[:-8]))
ret.append(curr[None])
if len(ret) < 2:
return self.standard_poses
return torch.cat(ret)
def get_pose_visual_input(self, subset="plain", std_pose=True, view_postfix="_1_front"):
keys = self.attr_keys[subset]
keys = keys[:min(len(keys), 8)]
all_froms, all_kpts, all_parses = [], [], []
all_from_kpts = []
for key in keys:
curr_key = key# + view_postfix
curr_from, curr_from_kpt, curr_parse = self.get_to_item(curr_key)
all_from_kpts += [curr_from_kpt]
curr_kpt = self.get_all_pose(key, std_pose=std_pose)
all_kpts += [curr_kpt]
all_froms += [curr_from.unsqueeze(0)]
all_parses += [curr_parse.unsqueeze(0)]
all_froms = torch.cat(all_froms)
all_parses = torch.cat(all_parses)
all_from_kpts = torch.cat(all_from_kpts)
return all_froms, all_parses, all_from_kpts, all_kpts #self.standard_poses
def get_attr_visual_input(self, subset="plain",view_postfix="_1_front"):
keys = self.attr_keys[subset]
keys = keys[:min(len(keys), 4)]
all_froms, all_parses, all_kpts = [], [], []
for key in keys:
curr_key = key# + view_postfix
curr_from, to_kpt, curr_parse = self.get_to_item(curr_key)
all_froms += [curr_from.unsqueeze(0)]
all_parses += [curr_parse.unsqueeze(0)]
all_kpts += [to_kpt.unsqueeze(0)]
all_froms = torch.cat(all_froms)
all_parses = torch.cat(all_parses)
all_kpts = torch.cat(all_kpts)
return all_froms, all_parses, all_kpts
def get_inputs_by_key(self, key):
#keys = self.attr_keys[subset]
#keys = keys[:min(len(keys), 4)]
curr_from, to_kpt, curr_parse = self.get_to_item(key)
return curr_from, curr_parse, to_kpt