add make code

make_openimages.py

@@ -0,0 +1,121 @@
+import json
+import os
+import pandas as pd
+from concurrent.futures import ThreadPoolExecutor
+from tqdm import tqdm
+
+from datasets import Dataset, load_dataset, Image
+import boto3
+
+
+def load_jsonl(file_path):
+    """
+    Loads a JSONL file and returns a list of Python dictionaries.
+    Each dictionary represents a JSON object from a line in the file.
+    """
+    data = []
+    with open(file_path, 'r', encoding='utf-8') as f:
+        for line in f:
+            try:
+                # Parse each line as a JSON object
+                json_object = json.loads(line.strip())
+                data.append(json_object)
+            except json.JSONDecodeError as e:
+                print(f"Error decoding JSON on line: {line.strip()} - {e}")
+    return data
+
+
+s3 = boto3.client('s3')
+
+
+def download_file(image_id, split):
+    bucket_name = "open-images-dataset"
+    object_key = f"{split}/{image_id}.jpg"
+    store_name = f"./open_images/images_files/{split}/{image_id}.jpg"
+    if os.path.exists(store_name):
+        return
+
+    try:
+        s3.download_file(bucket_name, object_key, store_name)
+    except KeyboardInterrupt:
+        raise
+    except:
+        print(f"Error getting {bucket_name}")
+
+    return
+
+
+def download_split(split):
+    workdir = "./open_images"
+
+    annot_file = os.path.join(workdir, f"{split}-images-with-rotation.csv")
+    df = pd.read_csv(annot_file)
+    df = pd.DataFrame(df)
+    os.makedirs(os.path.join(workdir, "images_files", split), exist_ok=True)
+    
+    # Prepare list of image IDs for multithreading
+    image_ids = df["ImageID"].tolist()
+    
+    # Use ThreadPoolExecutor with 6 threads and tqdm for progress
+    with ThreadPoolExecutor(max_workers=6) as executor:
+        try:
+            # Submit all download tasks
+            futures = [executor.submit(download_file, image_id, split) for image_id in image_ids]
+            
+            # Use tqdm to show progress
+            for future in tqdm(futures, desc=f"Downloading {split} images", unit="image"):
+                future.result()  # Wait for completion and handle any exceptions
+        except KeyboardInterrupt:
+            print("\nKeyboardInterrupt received. Shutting down threads...")
+            # Cancel all pending futures
+            for future in futures:
+                future.cancel()
+            # Shutdown the executor immediately
+            executor.shutdown(wait=False, cancel_futures=True)
+            print("All threads stopped.")
+            raise  # Re-raise to exit the program
+
+    return
+
+
+def test_dataset():
+    
+    return
+
+
+def prepare_subset(split):
+    workdir = "open_images"
+    csv_file = os.path.join(workdir, f"{split}-images-with-rotation.csv")
+    df = pd.read_csv(csv_file)
+    df = pd.DataFrame(df)
+    print(df)
+    
+    narratives = load_jsonl(os.path.join(workdir, "narratives", f"open_images_{split}_captions.jsonl"))
+    narratives = {x["image_id"]: x for x in narratives}
+    
+    def gen():
+        for _, row in df.iterrows():
+            item = {
+                "image_id": row["ImageID"],
+                "original_url": row["OriginalURL"],
+                "license": row["License"],
+                "author":  row["Author"],
+                "caption":  row["Title"],
+                "image_id": row["ImageID"],
+                "image": os.path.join(workdir, "images_files", split, row["ImageID"] + ".jpg"),
+                "narrative": narratives[row["ImageID"]]["caption"],
+            }
+            yield item
+
+    ds = Dataset.from_generator(gen)
+    ds = ds.cast_column("image", Image())
+    ds.save_to_disk(f"{workdir}/datasets/data/{split}", max_shard_size="400MB")
+    return
+
+
+if __name__ == "__main__":
+    # for split in ["train", "validation", "test"]:
+    #     download_split(split)
+    for split in ["validation"]:
+        prepare_subset(split)
+    # test_dataset()