Datasets:
Tasks:
Visual Question Answering
Formats:
parquet
Languages:
English
Size:
10K - 100K
ArXiv:
Tags:
medical
License:
flaviagiammarino
commited on
Commit
β’
647b9ea
1
Parent(s):
c9602cf
Upload process_dataset.py
Browse files- scripts/process_dataset.py +20 -55
scripts/process_dataset.py
CHANGED
@@ -1,70 +1,35 @@
|
|
1 |
-
"""This script
|
2 |
-
|
3 |
"""
|
4 |
|
5 |
-
import
|
|
|
6 |
import pickle
|
|
|
7 |
import pandas as pd
|
8 |
-
from PIL import Image
|
9 |
-
from tqdm import tqdm
|
10 |
|
11 |
-
# loop across the splits
|
12 |
for split in ["train", "val", "test"]:
|
13 |
|
|
|
|
|
14 |
# load the image-question-answer triplets
|
15 |
data = pd.DataFrame(pickle.load(open(f"pvqa/qas/{split}/{split}_qa.pkl", "rb")))
|
16 |
-
print(f"Total number of triplets in {split} set: {format(data.shape[0], ',.0f')}")
|
17 |
|
18 |
# drop the duplicate image-question-answer triplets
|
19 |
data = data.drop_duplicates(ignore_index=True)
|
20 |
-
print(f"Unique number of triplets in {split} set: {format(data.shape[0], ',.0f')}")
|
21 |
|
22 |
-
#
|
23 |
-
|
24 |
-
|
25 |
-
for image in tqdm(data["image"].unique()):
|
26 |
-
img = Image.open(f"pvqa/images/{split}/{image}.jpg")
|
27 |
-
byt = io.BytesIO()
|
28 |
-
img.save(byt, format="jpeg")
|
29 |
-
byt = byt.getvalue()
|
30 |
-
images[image] = {"path": None, "bytes": byt}
|
31 |
-
print(f"Unique number of images in {split} set: {format(len(images), ',.0f')}")
|
32 |
|
33 |
-
#
|
34 |
-
|
35 |
-
|
36 |
-
for _, row in data.iterrows():
|
37 |
-
dataset.append({
|
38 |
-
"image": images[row["image"]],
|
39 |
-
"question": row["question"],
|
40 |
-
"answer": row["answer"]
|
41 |
-
})
|
42 |
-
pd.DataFrame(dataset).to_parquet(f"data/{split}.parquet")
|
43 |
-
print("Done")
|
44 |
-
print("---------------------------------")
|
45 |
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
Done
|
54 |
-
---------------------------------
|
55 |
-
Total number of triplets in val set: 6,279
|
56 |
-
Unique number of triplets in val set: 6,259
|
57 |
-
Loading val set images...
|
58 |
-
100%|ββββββββββ| 832/832 [00:13<00:00, 59.49it/s]
|
59 |
-
Unique number of images in val set: 832
|
60 |
-
Writing data to data/val.parquet...
|
61 |
-
Done
|
62 |
-
---------------------------------
|
63 |
-
Total number of triplets in test set: 6,761
|
64 |
-
Unique number of triplets in test set: 6,719
|
65 |
-
Loading test set images...
|
66 |
-
100%|ββββββββββ| 858/858 [00:15<00:00, 53.93it/s]
|
67 |
-
Unique number of images in test set: 858
|
68 |
-
Writing data to data/test.parquet...
|
69 |
-
Done
|
70 |
-
'''
|
|
|
1 |
+
"""This script de-duplicates the data provided by the PathVQA authors,
|
2 |
+
creates an "imagefolder" dataset and pushes it to the hub.
|
3 |
"""
|
4 |
|
5 |
+
import os
|
6 |
+
import shutil
|
7 |
import pickle
|
8 |
+
import datasets
|
9 |
import pandas as pd
|
|
|
|
|
10 |
|
|
|
11 |
for split in ["train", "val", "test"]:
|
12 |
|
13 |
+
os.makedirs(f"data/{split}/", exist_ok=True)
|
14 |
+
|
15 |
# load the image-question-answer triplets
|
16 |
data = pd.DataFrame(pickle.load(open(f"pvqa/qas/{split}/{split}_qa.pkl", "rb")))
|
|
|
17 |
|
18 |
# drop the duplicate image-question-answer triplets
|
19 |
data = data.drop_duplicates(ignore_index=True)
|
|
|
20 |
|
21 |
+
# convert the image names to file names
|
22 |
+
data = data.rename(columns={"image": "file_name"})
|
23 |
+
data["file_name"] += ".jpg"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
25 |
+
# copy the images referenced by the question-answer pairs
|
26 |
+
for image in data["file_name"].unique():
|
27 |
+
shutil.copyfile(src=f"pvqa/images/{split}/{image}", dst=f"data/{split}/{image}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
28 |
|
29 |
+
# save the metadata
|
30 |
+
data.to_csv(f"data/{split}/metadata.csv", index=False)
|
31 |
+
|
32 |
+
# push the dataset to the hub
|
33 |
+
dataset = datasets.load_dataset("imagefolder", data_dir="data/")
|
34 |
+
dataset.push_to_hub("flaviagiammarino/path-vqa")
|
35 |
+
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|