diff --git a/.gitignore b/.gitignore index 97c323a..08b8396 100644 --- a/.gitignore +++ b/.gitignore @@ -8,3 +8,12 @@ /faiss_vector_db/faiss_index*/ /faiss_vector_db/WholeIngredientRecognition/faiss_index*/ /data_management/download_history.db +faiss_vector_db +*.jpg +*.png +*.pt +*.xlsx +SegFormer +*.bin +.vscode +.agents \ No newline at end of file diff --git a/data_management/count_delete_images.py b/data_management/count_delete_images.py new file mode 100644 index 0000000..d66108a --- /dev/null +++ b/data_management/count_delete_images.py @@ -0,0 +1,227 @@ +""" +统计每个类别的图片数量（包含train/val/test），并可选择删除样本不足的类别 +""" +import os +import shutil +from pathlib import Path +from collections import defaultdict + + +def count_images_by_class(dataset_root: str, output_file: str = None, delete_threshold: int = None, auto_confirm: bool = False): + """ + 统计每个类别的图片数量（包含train/val/test），并可选择删除样本不足的类别 + + Args: + dataset_root: 数据集根目录（包含train/val/test目录） + output_file: 输出文件路径（可选，如果提供则保存到文件） + delete_threshold: 删除阈值（可选，如果提供则删除总数少于该值的类别） + auto_confirm: 是否自动确认删除（默认False，需要用户确认） + """ + dataset_root = Path(dataset_root) + train_dir = dataset_root / 'train' + val_dir = dataset_root / 'val' + test_dir = dataset_root / 'test' + + if not train_dir.exists(): + print(f"错误: 训练目录不存在: {train_dir}") + return + + print("="*80) + print("每个类别的图片数量统计（包含train/val/test）") + print("="*80) + + class_counts = {} + class_counts_detail = {} # 详细统计：train/val/test分别的数量 + total_images = 0 + total_train = 0 + total_val = 0 + total_test = 0 + + # 获取所有类别文件夹（从train目录） + class_folders = sorted([f for f in train_dir.iterdir() if f.is_dir()]) + + # 也检查val和test目录中是否有train中没有的类别 + all_class_names = set() + for folder in class_folders: + all_class_names.add(folder.name) + + if val_dir.exists(): + for folder in val_dir.iterdir(): + if folder.is_dir(): + all_class_names.add(folder.name) + + if test_dir.exists(): + for folder in test_dir.iterdir(): + if folder.is_dir(): + all_class_names.add(folder.name) + + # 统计每个类别的图片数量 + for class_name in sorted(all_class_names): + train_count = 0 + val_count = 0 + test_count = 0 + + # 统计train目录 + train_class_dir = train_dir / class_name + if train_class_dir.exists(): + for ext in ['*.jpg', '*.jpeg', '*.png', '*.JPG', '*.JPEG', '*.PNG']: + train_count += len(list(train_class_dir.glob(ext))) + + # 统计val目录 + val_class_dir = val_dir / class_name + if val_class_dir.exists(): + for ext in ['*.jpg', '*.jpeg', '*.png', '*.JPG', '*.JPEG', '*.PNG']: + val_count += len(list(val_class_dir.glob(ext))) + + # 统计test目录 + test_class_dir = test_dir / class_name + if test_class_dir.exists(): + for ext in ['*.jpg', '*.jpeg', '*.png', '*.JPG', '*.JPEG', '*.PNG']: + test_count += len(list(test_class_dir.glob(ext))) + + total = train_count + val_count + test_count + class_counts[class_name] = total + class_counts_detail[class_name] = { + 'train': train_count, + 'val': val_count, + 'test': test_count, + 'total': total + } + total_images += total + total_train += train_count + total_val += val_count + total_test += test_count + + # 按数量排序 + sorted_classes = sorted(class_counts.items(), key=lambda x: x[1], reverse=True) + + # 修复文件开头的注释 + if output_file: + # 确保sorted_classes已定义 + pass + + # 打印结果 + print(f"\n总类别数: {len(class_counts)}") + print(f"总图片数: {total_images} (训练集: {total_train}, 验证集: {total_val}, 测试集: {total_test})") + print(f"平均每类: {total_images // len(class_counts) if len(class_counts) > 0 else 0} 张\n") + print("-"*100) + print(f"{'类别名称':<50} {'总计':<8} {'训练':<8} {'验证':<8} {'测试':<8} {'状态':<10}") + print("-"*100) + + # 统计样本不足的类别（假设阈值是10） + min_threshold = 10 + insufficient_count = 0 + + for class_name, count in sorted_classes: + detail = class_counts_detail[class_name] + status = "⚠样本不足" if count < min_threshold else "✓" + if count < min_threshold: + insufficient_count += 1 + print(f"{class_name:<50} {detail['total']:<8} {detail['train']:<8} {detail['val']:<8} {detail['test']:<8} {status:<10}") + + print("-"*100) + print(f"\n样本不足的类别数（<{min_threshold}张）: {insufficient_count}") + print(f"样本充足的类别数（>={min_threshold}张）: {len(class_counts) - insufficient_count}") + + # 删除样本不足的类别（如果指定了删除阈值） + if delete_threshold is not None: + classes_to_delete = [class_name for class_name, count in class_counts.items() if count < delete_threshold] + + if classes_to_delete: + print("\n" + "="*100) + print(f"发现 {len(classes_to_delete)} 个类别的总图片数少于 {delete_threshold} 张:") + print("="*100) + for class_name in classes_to_delete: + detail = class_counts_detail[class_name] + print(f" - {class_name}: {detail['total']} 张 (训练:{detail['train']}, 验证:{detail['val']}, 测试:{detail['test']})") + + # 确认删除 + if not auto_confirm: + confirm = input(f"\n是否删除这 {len(classes_to_delete)} 个类别? (yes/no): ").strip().lower() + if confirm not in ['yes', 'y']: + print("取消删除操作") + return + + # 执行删除 + deleted_count = 0 + for class_name in classes_to_delete: + try: + # 删除train目录中的类别文件夹 + train_class_dir = train_dir / class_name + if train_class_dir.exists(): + shutil.rmtree(train_class_dir) + print(f"✓ 已删除: {train_class_dir}") + + # 删除val目录中的类别文件夹 + val_class_dir = val_dir / class_name + if val_class_dir.exists(): + shutil.rmtree(val_class_dir) + print(f"✓ 已删除: {val_class_dir}") + + # 删除test目录中的类别文件夹 + test_class_dir = test_dir / class_name + if test_class_dir.exists(): + shutil.rmtree(test_class_dir) + print(f"✓ 已删除: {test_class_dir}") + + deleted_count += 1 + except Exception as e: + print(f"✗ 删除失败 {class_name}: {e}") + + print(f"\n删除完成: 成功删除 {deleted_count}/{len(classes_to_delete)} 个类别") + else: + print(f"\n没有找到总数少于 {delete_threshold} 张的类别") + + # 保存到文件（如果指定） + if output_file: + output_path = Path(output_file) + with open(output_path, 'w', encoding='utf-8') as f: + f.write("="*100 + "\n") + f.write("每个类别的图片数量统计（包含train/val/test）\n") + f.write("="*100 + "\n\n") + f.write(f"总类别数: {len(class_counts)}\n") + f.write(f"总图片数: {total_images} (训练集: {total_train}, 验证集: {total_val}, 测试集: {total_test})\n") + f.write(f"平均每类: {total_images // len(class_counts) if len(class_counts) > 0 else 0} 张\n\n") + f.write("-"*100 + "\n") + f.write(f"{'类别名称':<50} {'总计':<8} {'训练':<8} {'验证':<8} {'测试':<8} {'状态':<10}\n") + f.write("-"*100 + "\n") + + for class_name, count in sorted_classes: + detail = class_counts_detail[class_name] + status = "⚠样本不足" if count < min_threshold else "✓" + f.write(f"{class_name:<50} {detail['total']:<8} {detail['train']:<8} {detail['val']:<8} {detail['test']:<8} {status:<10}\n") + + f.write("-"*100 + "\n") + f.write(f"\n样本不足的类别数（<{min_threshold}张）: {insufficient_count}\n") + f.write(f"样本充足的类别数（>={min_threshold}张）: {len(class_counts) - insufficient_count}\n") + + print(f"\n✓ 统计结果已保存到: {output_path}") + + +if __name__ == "__main__": + import argparse + + parser = argparse.ArgumentParser(description='统计每个类别的图片数量，并可选择删除样本不足的类别') + parser.add_argument('--dataset', type=str, + default='dataset/WholeIngredientRecognition', + help='数据集根目录（默认: dataset/WholeIngredientRecognition）') + parser.add_argument('--output', type=str, default=None, + help='输出文件路径（可选，保存统计结果到文件）') + parser.add_argument('--delete', type=int, default=None, + help='删除阈值（可选，删除总数少于该值的类别，例如: --delete 20）') + parser.add_argument('--yes', action='store_true', + help='自动确认删除，不需要手动输入yes（谨慎使用）') + + args = parser.parse_args() + + # 获取项目根目录 + script_dir = Path(__file__).parent + project_root = script_dir.parent + + # 处理相对路径 + if not os.path.isabs(args.dataset): + dataset_path = project_root / args.dataset + else: + dataset_path = Path(args.dataset) + + count_images_by_class(str(dataset_path), args.output, args.delete, args.yes)