Upload cvdataset-layoutlmv3.py (#5)
Browse files- Upload cvdataset-layoutlmv3.py (99ec04f5ef5054baf038d8285bc4b84c189701b5)
- cvdataset-layoutlmv3.py +128 -0
cvdataset-layoutlmv3.py
ADDED
@@ -0,0 +1,128 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import os
|
3 |
+
from pathlib import Path
|
4 |
+
import datasets
|
5 |
+
from PIL import Image
|
6 |
+
import pandas as pd
|
7 |
+
|
8 |
+
logger = datasets.logging.get_logger(__name__)
|
9 |
+
_CITATION = """\
|
10 |
+
@article{LayoutLmv3 for CV extractions,
|
11 |
+
title={LayoutLmv3for Key Information Extraction},
|
12 |
+
author={MisaR&D Team},
|
13 |
+
year={2022},
|
14 |
+
}
|
15 |
+
"""
|
16 |
+
_DESCRIPTION = """\
|
17 |
+
CV is a collection of receipts. It contains, for each photo about cv personal, a list of OCRs - with the bounding box, text, and class. The goal is to benchmark "key information extraction" - extracting key information from documents
|
18 |
+
https://arxiv.org/abs/2103.14470
|
19 |
+
"""
|
20 |
+
|
21 |
+
def load_image(image_path):
|
22 |
+
image = Image.open(image_path)
|
23 |
+
w, h = image.size
|
24 |
+
return image, (w,h)
|
25 |
+
|
26 |
+
def normalize_bbox(bbox, size):
|
27 |
+
return [
|
28 |
+
int(1000 * bbox[0] / size[0]),
|
29 |
+
int(1000 * bbox[1] / size[1]),
|
30 |
+
int(1000 * bbox[2] / size[0]),
|
31 |
+
int(1000 * bbox[3] / size[1]),
|
32 |
+
]
|
33 |
+
|
34 |
+
def _get_drive_url(url):
|
35 |
+
base_url = 'https://drive.google.com/uc?id='
|
36 |
+
split_url = url.split('/')
|
37 |
+
|
38 |
+
return base_url + split_url[5]
|
39 |
+
|
40 |
+
_URLS = [
|
41 |
+
_get_drive_url("https://drive.google.com/file/d/1KdDBmGP96lFc7jv2Bf4eqrO121ST-TCh/"),
|
42 |
+
]
|
43 |
+
|
44 |
+
class DatasetConfig(datasets.BuilderConfig):
|
45 |
+
"""BuilderConfig for WildReceipt Dataset"""
|
46 |
+
def __init__(self, **kwargs):
|
47 |
+
"""BuilderConfig for WildReceipt Dataset.
|
48 |
+
Args:
|
49 |
+
**kwargs: keyword arguments forwarded to super.
|
50 |
+
"""
|
51 |
+
super(DatasetConfig, self).__init__(**kwargs)
|
52 |
+
|
53 |
+
|
54 |
+
class WildReceipt(datasets.GeneratorBasedBuilder):
|
55 |
+
BUILDER_CONFIGS = [
|
56 |
+
DatasetConfig(name="CV Extractions", version=datasets.Version("1.0.0"), description="CV dataset"),
|
57 |
+
]
|
58 |
+
|
59 |
+
def _info(self):
|
60 |
+
return datasets.DatasetInfo(
|
61 |
+
description=_DESCRIPTION,
|
62 |
+
features=datasets.Features(
|
63 |
+
{
|
64 |
+
"id": datasets.Value("string"),
|
65 |
+
"words": datasets.Sequence(datasets.Value("string")),
|
66 |
+
"bboxes": datasets.Sequence(datasets.Sequence(datasets.Value("int64"))),
|
67 |
+
"ner_tags": datasets.Sequence(
|
68 |
+
datasets.features.ClassLabel(
|
69 |
+
names=['person_name', 'dob_key', 'dob_value', 'gender_key', 'gender_value', 'phonenumber_key', 'phonenumber_value', 'email_key', 'email_value', 'address_key', 'address_value', 'socical_address_value', 'education', 'education_name', 'education_time', 'experience', 'experience_name', 'experience_time', 'information', 'undefined']
|
70 |
+
)
|
71 |
+
),
|
72 |
+
"image_path": datasets.Value("string"),
|
73 |
+
}
|
74 |
+
),
|
75 |
+
supervised_keys=None,
|
76 |
+
citation=_CITATION,
|
77 |
+
homepage="",
|
78 |
+
)
|
79 |
+
|
80 |
+
def _split_generators(self, dl_manager):
|
81 |
+
"""Returns SplitGenerators."""
|
82 |
+
"""Uses local files located with data_dir"""
|
83 |
+
downloaded_file = dl_manager.download_and_extract(_URLS)
|
84 |
+
dest = Path(downloaded_file[0])/'data1'
|
85 |
+
|
86 |
+
return [
|
87 |
+
datasets.SplitGenerator(
|
88 |
+
name=datasets.Split.TRAIN, gen_kwargs={"filepath": dest/"train.txt", "dest": dest}
|
89 |
+
),
|
90 |
+
datasets.SplitGenerator(
|
91 |
+
name=datasets.Split.TEST, gen_kwargs={"filepath": dest/"test.txt", "dest": dest}
|
92 |
+
),
|
93 |
+
]
|
94 |
+
|
95 |
+
def _generate_examples(self, filepath, dest):
|
96 |
+
|
97 |
+
df = pd.read_csv(dest/'class_list.txt', delimiter='\s', header=None)
|
98 |
+
id2labels = dict(zip(df[0].tolist(), df[1].tolist()))
|
99 |
+
|
100 |
+
|
101 |
+
logger.info("⏳ Generating examples from = %s", filepath)
|
102 |
+
|
103 |
+
item_list = []
|
104 |
+
with open(filepath, 'r') as f:
|
105 |
+
for line in f:
|
106 |
+
item_list.append(line.rstrip('\n\r'))
|
107 |
+
|
108 |
+
for guid, fname in enumerate(item_list):
|
109 |
+
|
110 |
+
data = json.loads(fname)
|
111 |
+
image_path = dest/data['file_name']
|
112 |
+
image, size = load_image(image_path)
|
113 |
+
boxes = [[i['box'][6], i['box'][7], i['box'][2], i['box'][3]] for i in data['annotations']]
|
114 |
+
|
115 |
+
text = [i['text'] for i in data['annotations']]
|
116 |
+
label = [id2labels[i['label']] for i in data['annotations']]
|
117 |
+
|
118 |
+
boxes = [normalize_bbox(box, size) for box in boxes]
|
119 |
+
|
120 |
+
flag=0
|
121 |
+
for i in boxes:
|
122 |
+
for j in i:
|
123 |
+
if j>1000:
|
124 |
+
flag+=1
|
125 |
+
pass
|
126 |
+
if flag>0: print(image_path)
|
127 |
+
|
128 |
+
yield guid, {"id": str(guid), "words": text, "bboxes": boxes, "ner_tags": label, "image_path": image_path}
|