-
Notifications
You must be signed in to change notification settings - Fork 1
/
processor_text.py
151 lines (131 loc) · 6.45 KB
/
processor_text.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 json
import torch
import clip
from PIL import Image
from tqdm import tqdm
from transformers import BertTokenizer, BertModel
import os
from transformers import BlipProcessor, BlipModel
import argparse
import datasets
def load_json(path):
f = open(path, 'r', encoding = 'utf-8')
data = json.load(f)
f.close()
return data
def json_save(data, path):
f = open(path, 'w', encoding = 'utf-8')
json.dump(data, f)
f.close()
def process_data(data, tokenizer, model_name, save_path):
fign2ocr = load_json('./data/fign2ocr_dic.json')
immage_preprocess_data = []
text_name_index = {}
text_name_index_2_subclass = {}
i = 0
count = 0
input_ids = []
attention_mask = []
text_token = []
ocr_input_ids = []
ocr_attention_mask = []
ocr_text_token = []
for line in tqdm(data):
caption = line['text']
file_name_index = line['file_name_index']
ocr = fign2ocr[file_name_index]
# inputs = tokenizer(caption, add_special_tokens=True,padding='max_length', truncation=True, max_length = 128, return_tensors = 'pt')
if model_name == 'BERT'or model_name in ['BLIP', 'BLIP-FLAN-T5-XXL', 'BLIP-FLAN-T5-XL']:
text_name_index_2_subclass[len(input_ids)] = line['class']
text_name_index[caption] = len(input_ids)
inputs = tokenizer(text = caption, add_special_tokens=True,padding='max_length', truncation=True, max_length = 128, return_tensors = 'pt')
input_ids.append(inputs['input_ids'])
attention_mask.append(inputs['attention_mask'])
inputs = tokenizer(text = ocr, add_special_tokens=True,padding='max_length', truncation=True, max_length = 128, return_tensors = 'pt')
ocr_input_ids.append(inputs['input_ids'])
ocr_attention_mask.append(inputs['attention_mask'])
elif model_name == 'CLIP':
text_name_index_2_subclass[len(text_token)] = line['class']
text_name_index[caption] = len(text_token)
inputs = tokenizer(caption, context_length=77, truncate = True)
inputs = inputs.squeeze(1)
text_token.append(inputs)
inputs = tokenizer(ocr, context_length=77, truncate = True)
inputs = inputs.squeeze(1)
ocr_text_token.append(inputs)
if model_name == 'BERT':
input_ids = torch.cat(input_ids)
attention_mask = torch.cat(attention_mask)
ocr_input_ids = torch.cat(ocr_input_ids)
ocr_attention_mask = torch.cat(ocr_attention_mask)
torch.save(input_ids, f'{save_path}input_ids.pt')
torch.save(attention_mask, f'{save_path}attention_mask.pt')
torch.save(ocr_input_ids, f'{save_path}ocr_input_ids.pt')
torch.save(ocr_attention_mask, f'{save_path}ocr_attention_mask.pt')
elif model_name == 'CLIP':
text_token = torch.cat(text_token)
torch.save(text_token, f'{save_path}text_mat.pt')
ocr_text_token = torch.cat(ocr_text_token)
torch.save(ocr_text_token, f'{save_path}ocr_text_mat.pt')
elif model_name in ['BLIP', 'BLIP-FLAN-T5-XXL', 'BLIP-FLAN-T5-XL']:
input_ids = torch.cat(input_ids)
attention_mask = torch.cat(attention_mask)
torch.save(input_ids, f'{save_path}input_ids.pt')
torch.save(attention_mask, f'{save_path}attention_mask.pt')
ocr_input_ids = torch.cat(ocr_input_ids)
ocr_attention_mask = torch.cat(ocr_attention_mask)
torch.save(ocr_input_ids, f'{save_path}ocr_input_ids.pt')
torch.save(ocr_attention_mask, f'{save_path}ocr_attention_mask.pt')
json_save(text_name_index, f'{save_path}text_index.json')
json_save(text_name_index_2_subclass, f'{save_path}text_name_index_2_subclass.json')
def run(args):
ds_remote = datasets.load_dataset("m-a-p/SciMMIR" )
valid_data = ds_remote['validation']
test_data = ds_remote['test']
print(args.model_name)
if args.model_name == 'BERT':
tokenizer = BertTokenizer.from_pretrained('bert-base-uncased' )
elif args.model_name == 'CLIP':
device = "cuda" if torch.cuda.is_available() else "cpu"
model, preprocess = clip.load("RN50x64", device=device)
tokenizer = clip.tokenize
elif args.model_name == 'BLIP':
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
processor = BlipProcessor.from_pretrained("Salesforce/blip-image-captioning-base" )
tokenizer = processor
elif args.model_name == 'BLIP-FLAN-T5-XL':
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
processor = BlipProcessor.from_pretrained("Salesforce/blip2-flan-t5-xl")
tokenizer = processor
elif args.model_name == 'BLIP-FLAN-T5-XXL':
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
processor = BlipProcessor.from_pretrained("Salesforce/blip2-flan-t5-xxl")
tokenizer = processor
if os.path.exists(args.save_path) == False:
os.mkdir(args.save_path)
#测试集也单独处理一份，这样可以节省验证的事件，验证的时候候选集小一点
if args.candidates_span == 'all_data':
test_data = ds_remote['test']
valid_data = ds_remote['validation']
train_data = ds_remote['train']
test_data = datasets.concatenate_datasets([test_data, valid_data, train_data])
elif args.candidates_span == 'test_split':
test_data = ds_remote['test']
data = test_data
save_path = f'{args.save_path}{args.model_name}_torch_data_test/'
if os.path.exists(save_path) == False:
os.mkdir(save_path)
process_data(data, tokenizer, args.model_name, save_path)
#验证集也单独处理一份，这样可以节省验证的事件，验证的时候候选集小一点
data = valid_data
save_path = f'{args.save_path}{args.model_name}_torch_data_valid/'
if os.path.exists(save_path) == False:
os.mkdir(save_path)
process_data(data, tokenizer, args.model_name, save_path)
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="config for process text")
parser.add_argument("--model_name" , type = str , default = "CLIP")
parser.add_argument("--save_path" , type = str , default = './data/')
parser.add_argument("--candidates_span" , type = str , default = "all_data")
args = parser.parse_args()
run(args)