Update scripts

scripts/dataset_utils.py

@@ -8,6 +8,8 @@ from urllib.parse import urlparse
 import requests
 import re
 from ast import literal_eval
+from PIL import Image
+import math
 from tqdm import tqdm
 tqdm.pandas()
 
@@ -340,4 +342,32 @@ def build_prompts(base_df, triplets, templates):
     triplets = triplets[['type','subject','question_for_image','question','possible_answers', 'relation', 's_uri', 'r_uri','a_uri','attribute','a_type']]
     return triplets
 
-    
+
+def resize_square(image, size=336, resample=Image.LANCZOS):
+    """
+    Resize an image to a square of the given size, first adding a black background if needed.
+    image:  a Pillow image instance
+    size:   an integer, the desired output size (width and height will be the same)
+    """
+    img_format = image.format
+    image = image.copy()
+    
+    size = [size, size]
+    img_size = image.size
+    ratio = min(size[0] / img_size[0], size[1] / img_size[1])
+    new_size = [
+        int(math.ceil(img_size[0] * ratio)),
+        int(math.ceil(img_size[1] * ratio))
+    ]
+    
+    image = image.resize((new_size[0], new_size[1]), resample)
+    
+    # Make the image square by adding black padding
+    max_dim = max(image.size)
+    new_img = Image.new("RGB", (max_dim, max_dim), (0, 0, 0))
+    new_img.paste(image, ((max_dim - image.size[0]) // 2, (max_dim - image.size[1]) // 2))
+    
+    # Resize to target size
+    # new_img = new_img.resize((size, size), resample)
+    new_img.format = img_format
+    return new_img

scripts/download_from_commons.py

@@ -0,0 +1,118 @@
+import pandas as pd
+import requests
+import os
+import argparse
+from tqdm import tqdm
+from  dataset_utils import resize_square
+from PIL import Image
+import cairosvg
+from io import BytesIO
+from dotenv import load_dotenv
+import time
+import random
+load_dotenv()
+
+DATA_DIR = os.environ['DATA_DIR']
+original_dir = os.path.join(DATA_DIR, 'original_path')
+resized_dir = os.path.join(DATA_DIR, 'resized_path')
+
+def fetch_image(url, max_retries=5):
+    headers = {
+        'User-Agent': random.choice([
+            'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36',
+            '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',
+            'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36',
+        ])
+    }
+    
+    for attempt in range(max_retries):
+        try:
+            response = requests.get(url, headers=headers, stream=True)
+            if response.status_code == 429:  # Too Many Requests
+                retry_after = int(response.headers.get("Retry-After", random.uniform(1, 5)))
+                time.sleep(retry_after)
+                continue
+            response.raise_for_status()
+            return response.content
+        except requests.exceptions.RequestException as e:
+            wait_time = 2 ** attempt + random.uniform(0, 1)  # Exponential backoff
+            print(f"Error fetching {url}: {e}, retrying in {wait_time:.2f}s")
+            time.sleep(wait_time)
+    
+    print(f"Failed to fetch {url} after {max_retries} attempts.")
+    return None
+
+def main(args):
+    df = pd.read_csv(args.csv)
+    df['original_path'] = ''
+    df['resized_path'] = ''
+
+    errors = 0
+    count = 0
+    
+    for index, row in tqdm(df.iterrows(), total=len(df)):
+        image_url = row['image']
+        subject = row['subject'].replace(' ', '_').replace('/', '_')
+        type_original_dir = os.path.join(original_dir, row['type'])
+        type_resized_dir = os.path.join(resized_dir, row['type'])
+        os.makedirs(type_original_dir, exist_ok=True)
+        os.makedirs(type_resized_dir, exist_ok=True)
+        
+        image_data = fetch_image(image_url)
+        if image_data:
+            try:
+                if image_url.endswith('.svg'):
+                    png_bytes = cairosvg.svg2png(bytestring=image_data)
+                    image = Image.open(BytesIO(png_bytes)).convert("RGBA")
+                else:
+                    image = Image.open(BytesIO(image_data))
+
+                if row['type'] == 'brands':
+                    image = image.convert("RGBA")
+                    white_bg = Image.new("RGB", image.size, (255, 255, 255))
+                    white_bg.paste(image, mask=image.split()[3])
+                    image = white_bg
+                
+                first_letter = subject[0].lower()
+                os.makedirs(os.path.join(type_original_dir, first_letter), exist_ok=True)
+                original_filepath = os.path.join(type_original_dir, first_letter, f'{subject}.jpg')
+                image.save(original_filepath, "JPEG", quality=95)
+
+                im = Image.open(original_filepath)
+                im = resize_square(im)
+                
+                os.makedirs(os.path.join(type_resized_dir, first_letter), exist_ok=True)
+                resized_filepath = os.path.join(type_resized_dir, first_letter, f'{subject}.jpg')
+                im.save(resized_filepath, im.format)
+
+                df.at[index, 'original_path'] = original_filepath if os.path.exists(original_filepath) else ''
+                df.at[index, 'resized_path'] = resized_filepath if os.path.exists(resized_filepath) else ''
+                
+            except Exception as e:
+                errors += 1
+                print(Exception(f"Failed to download image {subject}: {str(e)}"))
+                continue
+        else:
+            errors += 1
+            continue
+        count += 1
+
+    df = df[df['original_path'] != '']
+    df = df[df['resized_path'] != '']
+    df.to_csv(args.target_df, index=False)
+    
+    print(f'Finished downloading {count} images with {errors} errors')
+
+
+def get_exp_parser():
+    parser = argparse.ArgumentParser(add_help=False)
+    parser.add_argument('--base_df', type=str)
+    parser.add_argument('--target_df', type=str)
+    parser.add_argument('--width', type=int, default=336)
+    return parser
+
+
+if __name__ == "__main__":
+    parser = get_exp_parser()
+    args = parser.parse_args()
+    main(args)