/
flickr_dataset_All.py
270 lines (211 loc) · 9.71 KB
/
flickr_dataset_All.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
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
import torch
from torch.utils.data import Dataset
#import h5py
import json
import re
import os
from tsv_utils import *
import numpy as np
import nltk
#from flickr_utils import process_flickr_file
def calc_ious(ex_rois, gt_rois):
gt_rois = np.asarray(gt_rois, dtype=np.float32)
#print(gt_rois)
ex_area = (1. + ex_rois[:,2] - ex_rois[:,0]) * (1. + ex_rois[:,3] - ex_rois[:,1])
gt_area = (1. + gt_rois[:,2] - gt_rois[:,0]) * (1. + gt_rois[:,3] - gt_rois[:,1])
area_sum = ex_area.reshape((-1, 1)) + gt_area.reshape((1, -1))
lb = np.maximum(ex_rois[:,0].reshape((-1, 1)), gt_rois[:,0].reshape((1, -1)))
rb = np.minimum(ex_rois[:,2].reshape((-1, 1)), gt_rois[:,2].reshape((1, -1)))
tb = np.maximum(ex_rois[:,1].reshape((-1, 1)), gt_rois[:,1].reshape((1, -1)))
ub = np.minimum(ex_rois[:,3].reshape((-1, 1)), gt_rois[:,3].reshape((1, -1)))
width = np.maximum(1. + rb - lb, 0.)
height = np.maximum(1. + ub - tb, 0.)
area_i = width * height
area_u = area_sum - area_i
ious = area_i / area_u
result = np.where( ious >= 0.5)
listOfCordinates = list(zip(result[0], result[1]))
neg_res= np.where( ious <= 0.3)
neg_cord = list(zip(neg_res[0], neg_res[1]))
#print(listOfCordinates)
return ious, listOfCordinates, neg_cord
class CaptionDataset(Dataset):
def __init__(self, raw_data, tsv_data, w_vocab, srl_vocab, bio_vocab, mode, opt):#(self, data_json, tsv_data, word_dic, SRL_dic,img_fname, mode = 'train', sen_index = -1, flicker_all_cap = None,transform = None, max_len = 100, srl_training= False, bio_dic = None,label_dic= None):
self.raw_data = raw_data#json.load(open(opt.data_json, "r"))
self.sen_index = opt.sen_index
self.mode = mode
self.proposition_per_sen = 3
self.w_vocab = w_vocab
self.srl_vocab = srl_vocab
self.bio_vocab = bio_vocab
#self.max_len = opt.max_len
img_data = tsv_data
self.imgid2img = {}
for img_datum in img_data:
#print(img_datum['img_id'])
self.imgid2img[img_datum['img_id']+'.jpg'] = img_datum
#if mode!='train':
# Filter out the dataset
#img_fname = '/media/abhidip/2F1499756FA9B115/data/flickr/entity/flickr30k_entities-master/'
img_file = os.path.join(opt.data_path,self.mode+'.txt')
with open(img_file, 'r') as l:
lines = l.readlines()
img_list = [l.strip()+'.jpg' for l in lines]
data = []
for datum in self.raw_data:
image_name = datum['image']
if image_name in self.imgid2img and image_name in img_list:
#print(image_name)
data.append(datum)
self.frq_prop = []
self.data = self.process_data(data)
print(len(self.data))
print(len(data))
#print(self.data[0])
#print(self.data[1])
#print(self.data[2])
def __len__(self):
return len(self.data)
def process_single_sen(self, dsen, image):
p_data = []
p_tag =[]
sen = dsen["sentence"]
tags = dsen["tags"]
vn_tokens = dsen["vn_tokens"]
tokens = vn_tokens.split()#sen.split(" ")
tokens = [tok.lower().strip() for tok in tokens]
#tokens = nltk.word_tokenize(sen.lower())
#if len(tokens)>= self.max_len-2 :
#tokens = tokens[:self.max_len-2]
tokens = ['<start>'] + tokens + ['<end>']
enc_caption = [self.w_vocab(token) for token in tokens]
self.frq_prop.append(len(tags))
for i,tag_seq in enumerate(tags):
if self.proposition_per_sen == i:
break
tag = tag_seq["tag"].split(" ")
#if len(tag)>= self.max_len-2 :
#tag = tag[:self.max_len-2]
tag = ['<start>'] + tag + ['<end>']
#print("===sen:{}".format(tokens))
#print(len(enc_caption))
#print("====tag:{}".format(tag))
enc_tag = [self.bio_vocab(t) for t in tag]
#print(len(enc_tag))
assert len(enc_tag) == len(enc_caption)
for i in range(len(tag_seq['bboxes'])):
tag_seq['bboxes'][i]['en_srl'] = self.srl_vocab(tag_seq['bboxes'][i]['SRL'])#.get(tag_seq['bboxes'][i]['SRL'],def_srl)
#this_proposal = {''}
p_tag.append(enc_tag)
p_data.append(tag_seq['bboxes'])
return {'image':image, 'caption':sen, 'tokens': enc_caption, 'propositions':(p_data, p_tag) }
def process_data(self, old_data):
#sort(key=lambda x: len(x[1]), reverse = True)
data = []
if self.sen_index in range(5):
print("using {}th sentence from each image".format(self.sen_index))
else:
print("using all sentences")
for d in old_data:
if self.sen_index in range(5):
image = d["image"]
dsen = d["sentences"][self.sen_index]
#for dsen in d["sentences"]:
pdata = self.process_single_sen(dsen, image)
data.append(pdata)
else:
for dsen in d["sentences"]:
image = d["image"]
#for dsen in d["sentences"]:
pdata = self.process_single_sen(dsen, image)
#print("=========", pdata)
data.append(pdata)
return data
def process_box(self, roi_boxes, ac_boxes):
box_type = [self.srl_vocab('UNK')]*len(roi_boxes)
gt_boxes = [ac["box"] for ac in ac_boxes if len(ac["box"]) >0]
if len(gt_boxes) > 0:
ious, listOfCordinates, neg_cord = calc_ious(roi_boxes,gt_boxes)
for cord in listOfCordinates:
box_type[cord[0]] = ac_boxes[cord[1]]['en_srl']
for ncord in neg_cord:
if box_type[ncord[0]] == self.srl_vocab('UNK'):
box_type[ncord[0]] = self.srl_vocab('O')
return box_type
def __getitem__(self, index):
datum = self.data[index]
#print(datum)
#img features
image_name = datum['image']
img_info = self.imgid2img[image_name]
boxes = img_info['boxes'].copy()
feats = img_info['features'].copy()
#print(datum["bboxes"])
#print(datum["tagged_token"])
#print(datum["tag"].split(" "))
#print(datum["tokens"])
box_type = torch.zeros(self.proposition_per_sen,len(boxes)).long()
word_type = torch.zeros(self.proposition_per_sen, len(datum['tokens'])).long()
box_prop, word_prop = datum['propositions']
no_of_prop = len(box_prop)
for i,p in enumerate(box_prop):
box_type[i,:] = torch.LongTensor(self.process_box(boxes,p))
for i,p in enumerate(word_prop):
word_type[i,:] = torch.LongTensor(p)
# Normalize the boxes (to 0 ~ 1)
img_h, img_w = img_info['img_h'], img_info['img_w']
boxes[:, (0, 2)] /= img_w
boxes[:, (1, 3)] /= img_h
np.testing.assert_array_less(boxes, 1+1e-5)
np.testing.assert_array_less(-boxes, 0+1e-5)
#caplen = torch.LongTensor([datum["len"]])
box_type = torch.LongTensor(box_type)#.unsqueeze(dim=-1)
target = torch.Tensor(datum['tokens'])
feats = torch.FloatTensor(feats)
#all_captions = torch.LongTensor(self.caption_map[image_name])
return feats, torch.FloatTensor(boxes), box_type, target, index, no_of_prop, word_type
def collate_fn(data):
# Sort input data by decreasing lengths; why? apparent below
data.sort(key=lambda x: len(x[3]), reverse=True)
#print("data size {}".format(len(data)))
#print(data)
images, boxes, box_type, captions, ids, no_of_prop, w_type = zip(*data)
images = torch.stack(images, 0)
boxes = torch.stack(boxes, 0)
box_type = torch.stack(box_type,0)
lengths = [len(cap) for cap in captions]
targets = torch.zeros(len(captions), max(lengths)).long()
w_type_padded = torch.zeros(len(captions), len(w_type[0]), max(lengths)).long()
for i,cap in enumerate(captions):
end = lengths[i]
targets[i, :end] = cap[:end]
w_type_padded[i,:,:end] = w_type[i][:end]
#print(caplen)
return images, boxes, box_type, targets, lengths, ids, no_of_prop, w_type_padded
def get_dataloader(train_data, val_data, opt):
train_loader = torch.utils.data.DataLoader(
train_data,
batch_size=opt.batch_size, shuffle=True, num_workers=opt.workers, collate_fn = collate_fn, pin_memory=True)
val_loader = torch.utils.data.DataLoader(
val_data,
batch_size=opt.batch_size, shuffle=False, num_workers=opt.workers, collate_fn=collate_fn, pin_memory=True)
return train_loader, val_loader
def get_test_loader(test_data, opt):
test_loader = torch.utils.data.DataLoader(
test_data,
batch_size=opt.batch_size, shuffle=False, num_workers=opt.workers, collate_fn=collate_fn, pin_memory=True)
return test_loader
if __name__ == '__main__':
datafile = '/media/abhidip/2F1499756FA9B115/data/flickr/abhidip_splits/flickrdata_VN_SRL_martha_BBOX.json'
tsv_path = '/media/abhidip/2F1499756FA9B115/data/flickr/image_feat/'
flickr_file = '/media/abhidip/2F1499756FA9B115/data/flickr/flickr30k/results_20130124.token'
img_data = []
tsv_path1 = os.path.join(tsv_path, "train.tsv")
tsv_path2 = os.path.join(tsv_path, "dev.tsv")
tsv_path3 = os.path.join(tsv_path, "test.tsv")
img_data.extend(load_obj_tsv(tsv_path1, topk=None))
img_data.extend(load_obj_tsv(tsv_path2, topk=None))
img_data.extend(load_obj_tsv(tsv_path3, topk=None))
dataset = CaptionDataset(datafile, img_data, 'word_dic_A.txt', 'srl_dic_A.txt', mode = 'train',flicker_all_cap = flickr_file)
img, boxes, box_type, caption, caplen = dataset[9751]
print(box_type)