split_dataset.py

"""
数据集划分脚本
将 dataset/trashnet/ 按比例拆分为 train / val / test
生成结构:
  dataset/
  ├── trashnet/         (原始数据, 不动)
  ├── train/
  │   ├── cardboard/
  │   └── ...
  ├── val/
  │   ├── cardboard/
  │   └── ...
  └── test/
      ├── cardboard/
      └── ...
"""

import argparse
import random
import shutil
from pathlib import Path


def split_dataset(data_dir, output_dir, train_ratio=0.7, val_ratio=0.15, seed=42):
    data_dir = Path(data_dir)
    output_dir = Path(output_dir)

    if not data_dir.exists():
        print(f"✗ 数据集路径不存在: {data_dir}")
        return

    random.seed(seed)

    # 收集所有类别
    classes = sorted([d.name for d in data_dir.iterdir() if d.is_dir()])
    print(f"发现 {len(classes)} 个类别: {classes}")

    splits = {"train": train_ratio, "val": val_ratio, "test": 1 - train_ratio - val_ratio}
    print(f"\n划分比例: {splits}")

    for cls in classes:
        src_dir = data_dir / cls
        images = sorted([f for f in src_dir.iterdir() if f.is_file()])
        random.shuffle(images)

        n = len(images)
        n_train = int(n * train_ratio)
        n_val = int(n * val_ratio)

        split_files = {
            "train": images[:n_train],
            "val": images[n_train:n_train + n_val],
            "test": images[n_train + n_val:],
        }

        for split_name, files in split_files.items():
            dest_dir = output_dir / split_name / cls
            dest_dir.mkdir(parents=True, exist_ok=True)
            for f in files:
                shutil.copy2(f, dest_dir / f.name)

        print(f"  {cls:12s}: train={len(split_files['train']):4d}  "
              f"val={len(split_files['val']):4d}  "
              f"test={len(split_files['test']):4d}")

    print(f"\n✓ 划分完成!")
    print(f"  输出目录: {output_dir.resolve()}")
    print(f"  结构: ")
    print(f"    {output_dir.name}/")
    for split_name in ["train", "val", "test"]:
        total = sum(len(list((output_dir / split_name / cls).iterdir())) for cls in classes)
        print(f"    ├── {split_name}/  ({total} 张)")


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="划分训练集/验证集/测试集")
    parser.add_argument("--data-dir", default="dataset/trashnet", help="原始数据集路径")
    parser.add_argument("--output-dir", default="dataset", help="输出目录 (将在其中创建 train/val/test)")
    parser.add_argument("--train-ratio", type=float, default=0.7, help="训练集比例")
    parser.add_argument("--val-ratio", type=float, default=0.15, help="验证集比例")
    parser.add_argument("--seed", type=int, default=42, help="随机种子")
    args = parser.parse_args()
    split_dataset(args.data_dir, args.output_dir, args.train_ratio, args.val_ratio, args.seed)