-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathdata_loader.py
163 lines (143 loc) · 5.89 KB
/
data_loader.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
import json
import jsonlines
import os
from nltk.tokenize import RegexpTokenizer
class BLEUDataLoader:
def __init__(self,
ref_path="../captions/ref/gpt4v_llava_10k_test.json", ref_repeat=4,
hypo_dir="../captions/hypo", split=True) -> None:
self.ref_path = ref_path
self.ref_repeat = ref_repeat
self.hypo_dir = hypo_dir
self.split = split
self.ref = self.set_ref()
self.hypo_dict = self.set_hypo_dict()
def corpus_process(self, paragraph):
if self.split == False:
return paragraph
tokenizer = RegexpTokenizer(r'\w+')
pure_words = tokenizer.tokenize(paragraph)
return pure_words
def set_ref(self):
# [[ref_a1],]
ref = []
with open(self.ref_path, 'r') as f:
content = json.load(f)
for i in range(int(len(content)/self.ref_repeat)):
ref.append([self.corpus_process(content[self.ref_repeat*i]["conversations"][1]["value"])])
return ref
def gen_hypo(self, hypo_path="../captions/hypo/output-caption-gpt4v-hypo-0.01.jsonl"):
# [hypo_a,]
hypo = []
with open(hypo_path, 'r') as f:
for chat in jsonlines.Reader(f):
hypo.append(self.corpus_process(chat['outputs']))
return hypo
def set_hypo_dict(self):
# {filename_attr: [hypo_a,]}
hypo_dict = {}
for file_name in os.listdir(self.hypo_dir):
hypo_path = os.path.join(self.hypo_dir, file_name)
filename_attr = file_name[21:-6]
hypo = self.gen_hypo(hypo_path)
hypo_dict[filename_attr] = hypo
return hypo_dict
class CIDErDataLoader:
def __init__(self,
ref_path="../captions/ref/gpt4v_llava_10k_test.json", ref_repeat=4,
hypo_dir="../captions/hypo", split=True) -> None:
self.ref_path = ref_path
self.ref_repeat = ref_repeat
self.hypo_dir = hypo_dir
self.split = split
self.img = self.set_img()
self.ref = self.set_ref()
self.hypo_dict = self.set_hypo_dict()
def corpus_process(self, paragraph):
if self.split == False:
return paragraph
tokenizer = RegexpTokenizer(r'\w+')
pure_words = tokenizer.tokenize(paragraph)
return pure_words
def set_img(self):
img = []
with open(self.ref_path, 'r') as f:
content = json.load(f)
for i in range(int(len(content)/self.ref_repeat)):
img.append(int(content[self.ref_repeat*i]["image"].split('.')[0]))
return img
def set_ref(self):
# {<image>: [<tokenized reference sentence>]}
ref = {}
with open(self.ref_path, 'r') as f:
content = json.load(f)
for i in range(int(len(content)/self.ref_repeat)):
ref[self.img[len(ref)]] = [self.corpus_process(content[self.ref_repeat*i]["conversations"][1]["value"])]
return ref
def gen_hypo(self, hypo_path="../captions/hypo/output-caption-gpt4v-hypo-0.01.jsonl"):
# {<image>: [<tokenized hypothesis sentence>]}
hypo = {}
with open(hypo_path, 'r') as f:
for chat in jsonlines.Reader(f):
hypo[self.img[len(hypo)]] = [self.corpus_process(chat['outputs'])]
return hypo
def set_hypo_dict(self):
# {filename_attr: {<image>: <tokenized hypothesis sentence>}}
hypo_dict = {}
for file_name in os.listdir(self.hypo_dir):
hypo_path = os.path.join(self.hypo_dir, file_name)
filename_attr = file_name[21:-6]
hypo = self.gen_hypo(hypo_path)
hypo_dict[filename_attr] = hypo
return hypo_dict
class CIDErDDataLoader:
def __init__(self,
ref_path="../captions/ref/gpt4v_llava_10k_test.json", ref_repeat=4,
hypo_dir="../captions/hypo", split=True) -> None:
self.ref_path = ref_path
self.ref_repeat = ref_repeat
self.hypo_dir = hypo_dir
self.split = split
self.img = self.set_img()
self.ref = self.set_ref()
self.hypo_dict = self.set_hypo_dict()
def corpus_process(self, paragraph):
if self.split == False:
return paragraph
tokenizer = RegexpTokenizer(r'\w+')
pure_words = tokenizer.tokenize(paragraph)
return pure_words
def set_img(self):
img = []
with open(self.ref_path, 'r') as f:
content = json.load(f)
for i in range(int(len(content)/self.ref_repeat)):
img.append(int(content[self.ref_repeat*i]["image"].split('.')[0]))
return img
def set_ref(self):
# {<image>: <tokenized reference sentence>}
ref = {}
with open(self.ref_path, 'r') as f:
content = json.load(f)
for i in range(int(len(content)/self.ref_repeat)):
ref[self.img[len(ref)]] = self.corpus_process(content[self.ref_repeat*i]["conversations"][1]["value"])
return ref
def gen_hypo(self, hypo_path="../captions/hypo/output-caption-gpt4v-hypo-0.01.jsonl"):
# {"image_id": <image>, "caption": <tokenized hypothesis sentence>}
hypo = []
with open(hypo_path, 'r') as f:
for chat in jsonlines.Reader(f):
res = {}
res["image_id"] = self.img[len(hypo)]
res["caption"] = self.corpus_process(chat['outputs'])
hypo.append(res)
return hypo
def set_hypo_dict(self):
# {filename_attr: {"image_id": <image>, "caption": <tokenized hypothesis sentence>}}
hypo_dict = {}
for file_name in os.listdir(self.hypo_dir):
hypo_path = os.path.join(self.hypo_dir, file_name)
filename_attr = file_name[21:-6]
hypo = self.gen_hypo(hypo_path)
hypo_dict[filename_attr] = hypo
return hypo_dict