File size: 2,030 Bytes
3e1d9f3 |
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 |
from ..root import (
DATASETS,
QUESTION_PLACEHOLDER,
IMAGE_PLACEHOLDER,
BOXES_PLACEHOLDER,
)
from ..utils import MInstrDataset
from ..utils.flickr30k_entities_utils import PHRASE_ST_PLACEHOLDER, PHRASE_ED_PLACEHOLDER
@DATASETS.register_module()
class GPT4Gen(MInstrDataset):
def __init__(self, *args, version, **kwargs):
super().__init__(*args, **kwargs, placeholders=(IMAGE_PLACEHOLDER, QUESTION_PLACEHOLDER))
self.version = version
assert version in ['a', 'c', 'bc']
def __getitem__(self, item):
raw = self.get_raw_item(item)
#
image = self.get_image(raw['img_path'])
#
boxes = raw['boxes']
#
question = raw['question']
question = question.replace(PHRASE_ST_PLACEHOLDER, '').replace(PHRASE_ED_PLACEHOLDER, BOXES_PLACEHOLDER)
final_question = self.get_template().replace(QUESTION_PLACEHOLDER, question)
query_boxes_seq = raw['question_boxes_seq']
if self.version == 'a':
final_answer = raw['answer']
answer_boxes_seq = None
elif self.version == 'c':
final_answer = raw['cot_with_ans'].replace(PHRASE_ST_PLACEHOLDER, '').replace(PHRASE_ED_PLACEHOLDER, '')
answer_boxes_seq = None
elif self.version == 'bc':
final_answer = raw['cot_with_ans'].replace(PHRASE_ST_PLACEHOLDER, '').replace(PHRASE_ED_PLACEHOLDER, BOXES_PLACEHOLDER)
answer_boxes_seq = raw['answer_boxes_seq']
else:
assert False
ret = {
'image': image,
'target': {'boxes': boxes},
'conversations': [
{
'from': 'human',
'value': final_question,
'boxes_seq': query_boxes_seq,
},
{
'from': 'gpt',
'value': final_answer,
'boxes_seq': answer_boxes_seq,
}
]
}
return ret
|