[tool] Add pixelart dataset downloader

Reference: https://huggingface.co/datasets/jainr3/diffusiondb-pixelart

Author: OleehyO

tools/pixelart_dataset/download_convert.py

@@ -0,0 +1,58 @@
+import os
+import json
+from datasets import load_dataset
+
+
+def generate_sequential_filename(index):
+    """Generate sequential filename with leading zeros (00001.png, 00002.png, etc.)."""
+    return f"{index + 1:05d}.png"
+
+
+def setup_directories(base_dir):
+    """Create train directory if it doesn't exist."""
+    train_dir = os.path.join(base_dir, "train")
+    os.makedirs(train_dir, exist_ok=True)
+    return train_dir
+
+
+def main():
+    # Setup directories
+    base_dir = os.path.dirname(os.path.abspath(__file__))
+    train_dir = setup_directories(base_dir)
+
+    # Load the dataset
+    print("Loading diffusiondb-pixelart dataset (2k_all subset)...")
+    dataset = load_dataset("jainr3/diffusiondb-pixelart", "2k_all")
+
+    # Get all data
+    data = dataset["train"]
+    num_samples = len(data)
+
+    # Process all data as train data
+    train_metadata = []
+    print(f"Processing all {num_samples} samples for training...")
+
+    for idx in range(num_samples):
+        item = data[idx]
+        prompt = item["text"]
+        image = item["image"]
+
+        filename = generate_sequential_filename(idx)
+        save_path = os.path.join(train_dir, filename)
+
+        # Save the image
+        image.save(save_path)
+
+        train_metadata.append({"file_name": filename, "prompt": prompt})
+
+    # Save train metadata
+    with open(os.path.join(train_dir, "metadata.jsonl"), "w", encoding="utf-8") as f:
+        for item in train_metadata:
+            f.write(json.dumps(item) + "\n")
+
+    print("Conversion complete!")
+    print(f"Total samples: {len(train_metadata)}")
+
+
+if __name__ == "__main__":
+    main()