Datasets:
Tasks:
Visual Question Answering
Modalities:
Image
Formats:
imagefolder
Languages:
English
Size:
1K - 10K
License:
Update scripts
Browse files- scripts/dataset_utils.py +31 -1
- scripts/download_from_commons.py +118 -0
scripts/dataset_utils.py
CHANGED
@@ -8,6 +8,8 @@ from urllib.parse import urlparse
|
|
8 |
import requests
|
9 |
import re
|
10 |
from ast import literal_eval
|
|
|
|
|
11 |
from tqdm import tqdm
|
12 |
tqdm.pandas()
|
13 |
|
@@ -340,4 +342,32 @@ def build_prompts(base_df, triplets, templates):
|
|
340 |
triplets = triplets[['type','subject','question_for_image','question','possible_answers', 'relation', 's_uri', 'r_uri','a_uri','attribute','a_type']]
|
341 |
return triplets
|
342 |
|
343 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
import requests
|
9 |
import re
|
10 |
from ast import literal_eval
|
11 |
+
from PIL import Image
|
12 |
+
import math
|
13 |
from tqdm import tqdm
|
14 |
tqdm.pandas()
|
15 |
|
|
|
342 |
triplets = triplets[['type','subject','question_for_image','question','possible_answers', 'relation', 's_uri', 'r_uri','a_uri','attribute','a_type']]
|
343 |
return triplets
|
344 |
|
345 |
+
|
346 |
+
def resize_square(image, size=336, resample=Image.LANCZOS):
|
347 |
+
"""
|
348 |
+
Resize an image to a square of the given size, first adding a black background if needed.
|
349 |
+
image: a Pillow image instance
|
350 |
+
size: an integer, the desired output size (width and height will be the same)
|
351 |
+
"""
|
352 |
+
img_format = image.format
|
353 |
+
image = image.copy()
|
354 |
+
|
355 |
+
size = [size, size]
|
356 |
+
img_size = image.size
|
357 |
+
ratio = min(size[0] / img_size[0], size[1] / img_size[1])
|
358 |
+
new_size = [
|
359 |
+
int(math.ceil(img_size[0] * ratio)),
|
360 |
+
int(math.ceil(img_size[1] * ratio))
|
361 |
+
]
|
362 |
+
|
363 |
+
image = image.resize((new_size[0], new_size[1]), resample)
|
364 |
+
|
365 |
+
# Make the image square by adding black padding
|
366 |
+
max_dim = max(image.size)
|
367 |
+
new_img = Image.new("RGB", (max_dim, max_dim), (0, 0, 0))
|
368 |
+
new_img.paste(image, ((max_dim - image.size[0]) // 2, (max_dim - image.size[1]) // 2))
|
369 |
+
|
370 |
+
# Resize to target size
|
371 |
+
# new_img = new_img.resize((size, size), resample)
|
372 |
+
new_img.format = img_format
|
373 |
+
return new_img
|
scripts/download_from_commons.py
ADDED
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import pandas as pd
|
2 |
+
import requests
|
3 |
+
import os
|
4 |
+
import argparse
|
5 |
+
from tqdm import tqdm
|
6 |
+
from dataset_utils import resize_square
|
7 |
+
from PIL import Image
|
8 |
+
import cairosvg
|
9 |
+
from io import BytesIO
|
10 |
+
from dotenv import load_dotenv
|
11 |
+
import time
|
12 |
+
import random
|
13 |
+
load_dotenv()
|
14 |
+
|
15 |
+
DATA_DIR = os.environ['DATA_DIR']
|
16 |
+
original_dir = os.path.join(DATA_DIR, 'original_path')
|
17 |
+
resized_dir = os.path.join(DATA_DIR, 'resized_path')
|
18 |
+
|
19 |
+
def fetch_image(url, max_retries=5):
|
20 |
+
headers = {
|
21 |
+
'User-Agent': random.choice([
|
22 |
+
'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36',
|
23 |
+
'Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36',
|
24 |
+
'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36',
|
25 |
+
])
|
26 |
+
}
|
27 |
+
|
28 |
+
for attempt in range(max_retries):
|
29 |
+
try:
|
30 |
+
response = requests.get(url, headers=headers, stream=True)
|
31 |
+
if response.status_code == 429: # Too Many Requests
|
32 |
+
retry_after = int(response.headers.get("Retry-After", random.uniform(1, 5)))
|
33 |
+
time.sleep(retry_after)
|
34 |
+
continue
|
35 |
+
response.raise_for_status()
|
36 |
+
return response.content
|
37 |
+
except requests.exceptions.RequestException as e:
|
38 |
+
wait_time = 2 ** attempt + random.uniform(0, 1) # Exponential backoff
|
39 |
+
print(f"Error fetching {url}: {e}, retrying in {wait_time:.2f}s")
|
40 |
+
time.sleep(wait_time)
|
41 |
+
|
42 |
+
print(f"Failed to fetch {url} after {max_retries} attempts.")
|
43 |
+
return None
|
44 |
+
|
45 |
+
def main(args):
|
46 |
+
df = pd.read_csv(args.csv)
|
47 |
+
df['original_path'] = ''
|
48 |
+
df['resized_path'] = ''
|
49 |
+
|
50 |
+
errors = 0
|
51 |
+
count = 0
|
52 |
+
|
53 |
+
for index, row in tqdm(df.iterrows(), total=len(df)):
|
54 |
+
image_url = row['image']
|
55 |
+
subject = row['subject'].replace(' ', '_').replace('/', '_')
|
56 |
+
type_original_dir = os.path.join(original_dir, row['type'])
|
57 |
+
type_resized_dir = os.path.join(resized_dir, row['type'])
|
58 |
+
os.makedirs(type_original_dir, exist_ok=True)
|
59 |
+
os.makedirs(type_resized_dir, exist_ok=True)
|
60 |
+
|
61 |
+
image_data = fetch_image(image_url)
|
62 |
+
if image_data:
|
63 |
+
try:
|
64 |
+
if image_url.endswith('.svg'):
|
65 |
+
png_bytes = cairosvg.svg2png(bytestring=image_data)
|
66 |
+
image = Image.open(BytesIO(png_bytes)).convert("RGBA")
|
67 |
+
else:
|
68 |
+
image = Image.open(BytesIO(image_data))
|
69 |
+
|
70 |
+
if row['type'] == 'brands':
|
71 |
+
image = image.convert("RGBA")
|
72 |
+
white_bg = Image.new("RGB", image.size, (255, 255, 255))
|
73 |
+
white_bg.paste(image, mask=image.split()[3])
|
74 |
+
image = white_bg
|
75 |
+
|
76 |
+
first_letter = subject[0].lower()
|
77 |
+
os.makedirs(os.path.join(type_original_dir, first_letter), exist_ok=True)
|
78 |
+
original_filepath = os.path.join(type_original_dir, first_letter, f'{subject}.jpg')
|
79 |
+
image.save(original_filepath, "JPEG", quality=95)
|
80 |
+
|
81 |
+
im = Image.open(original_filepath)
|
82 |
+
im = resize_square(im)
|
83 |
+
|
84 |
+
os.makedirs(os.path.join(type_resized_dir, first_letter), exist_ok=True)
|
85 |
+
resized_filepath = os.path.join(type_resized_dir, first_letter, f'{subject}.jpg')
|
86 |
+
im.save(resized_filepath, im.format)
|
87 |
+
|
88 |
+
df.at[index, 'original_path'] = original_filepath if os.path.exists(original_filepath) else ''
|
89 |
+
df.at[index, 'resized_path'] = resized_filepath if os.path.exists(resized_filepath) else ''
|
90 |
+
|
91 |
+
except Exception as e:
|
92 |
+
errors += 1
|
93 |
+
print(Exception(f"Failed to download image {subject}: {str(e)}"))
|
94 |
+
continue
|
95 |
+
else:
|
96 |
+
errors += 1
|
97 |
+
continue
|
98 |
+
count += 1
|
99 |
+
|
100 |
+
df = df[df['original_path'] != '']
|
101 |
+
df = df[df['resized_path'] != '']
|
102 |
+
df.to_csv(args.target_df, index=False)
|
103 |
+
|
104 |
+
print(f'Finished downloading {count} images with {errors} errors')
|
105 |
+
|
106 |
+
|
107 |
+
def get_exp_parser():
|
108 |
+
parser = argparse.ArgumentParser(add_help=False)
|
109 |
+
parser.add_argument('--base_df', type=str)
|
110 |
+
parser.add_argument('--target_df', type=str)
|
111 |
+
parser.add_argument('--width', type=int, default=336)
|
112 |
+
return parser
|
113 |
+
|
114 |
+
|
115 |
+
if __name__ == "__main__":
|
116 |
+
parser = get_exp_parser()
|
117 |
+
args = parser.parse_args()
|
118 |
+
main(args)
|