text
stringlengths 0
237
|
|---|
# Cops-Ref Dataset
|
This txt file gives explanations for the annotation format of Cops-Ref (refGQA_v0.json).
|
The annotation format is similar to that of MattNet (Yu et al., CVPR 2018). The annotation dictionary contains 9 keys, including refs, images, anns, cat_to_ix, word_to_ix, label_length, sentences, att_to_ix and att_to_cnt.
|
### refs
|
Every element in refs has the following keys
|
refs[1000]
|
{
|
valid_flag: 1 or -1 # those expression with -1 as valid flag will be discareded.
|
split: 'train', 'val' or 'test'
|
name: 'woman', 'napkin' # object category name
|
box: [x, y, h, w]
|
ann_id: 1, 2, ... # index for the referred region
|
image_id: '2370370', '2406655' # key for the referred image
|
logic: 'chain' # main logic of the expressioni
|
ref_id: index for the referring expression
|
neg_img_list:
|
{
|
'catAndcat': ['2325119', '2406957', '2320024'],
|
'random': ['2358081', '2394889', '2361775'],
|
'catAndAttr': ['2361313', '2384329', '2390550'],
|
'cat': ['2361313', '2352441', '2390550']
|
} # including the index of the distracting image
|
sentences: [] # a list of sentence info, including tokens, sent_id, sent
|
...
|
}
|
### images
|
image[1000]
|
{
|
height: 375
|
width: 500
|
image_id: '2327364'
|
ref_ids: [142897, 164127] # referring expressions assoicated with the image
|
ann_ids: [46640, 46642, 46643, 46644, 46645, 46646, 46647, 46649, 282639, 377642, 377643, 377644, 377650, 888232, 888253, 1382479, 1382480, 1382481, 1382482, 1382483, 1382484, 1382485] # indexes for regions associated with the images
|
...
|
}
|
### anns
|
anns[1000]
|
{
|
image_id: '2412166'
|
box: [291, 114, 14, 11]
|
category_id: 171 # index for the category of the region
|
ann_id: 1000
|
}
|
### sentences
|
{
|
'tokens': # list of words
|
'sent_id': # index for the sentence
|
...
|
}
|
### cat_to_ix
|
mapping category to index
|
### word_to_ix
|
mapping word to index
|
### att_to_ix
|
mapping attributes to index
|
### att_to_cnt
|
storing counts of most frequent attributes
|
### label_length
|
maxinum length of the expressions in the baselines
|
No dataset card yet
- Downloads last month
- 25