#!/usr/bin/env python3
"""
检查fast_results.pkl文件的结构和内容
显示评估结果的统计信息
"""

import os
import sys
import pickle

def load_results_sample(file_path, max_samples=5):
    """安全地加载结果文件的一部分"""
    print(f"📂 分析文件: {file_path}")

    # 检查文件大小
    file_size = os.path.getsize(file_path)
    print(f"📏 文件大小: {file_size / (1024**3):.2f} GB")
    if file_size > 50 * 1024**3:  # 50GB
        print("⚠️ 文件过大，将只分析前几个样本")

    try:
        with open(file_path, 'rb') as f:
            # 尝试加载
            results = pickle.load(f)

        print(f"✅ 成功加载数据")
        print(f"📊 数据类型: {type(results)}")

        if isinstance(results, list):
            analyze_list_results(results, max_samples)
        else:
            print(f"⚠️ 意外的数据类型: {type(results)}")

    except Exception as e:
        print(f"❌ 加载失败: {e}")
        return None

def analyze_list_results(results, max_samples=5):
    """分析列表类型的评估结果"""
    total_samples = len(results)
    print(f"📏 总样本数: {total_samples}")

    if total_samples == 0:
        print("⚠️ 结果列表为空")
        return

    # 分析前几个样本
    samples_to_analyze = min(max_samples, total_samples)
    print(f"🔍 分析前 {samples_to_analyze} 个样本:")

    sample_stats = {
        'has_boxes_3d': 0,
        'has_scores_3d': 0,
        'has_labels_3d': 0,
        'has_masks_bev': 0,
        'total_boxes': 0,
        'box_score_ranges': [],
        'mask_shapes': []
    }

    for i in range(samples_to_analyze):
        sample = results[i]
        print(f"\n  样本 {i+1}:")

        if isinstance(sample, dict):
            # 检查3D检测结果
            if 'boxes_3d' in sample:
                sample_stats['has_boxes_3d'] += 1
                boxes = sample['boxes_3d']
                if hasattr(boxes, 'tensor'):
                    num_boxes = boxes.tensor.shape[0]
                else:
                    num_boxes = len(boxes) if hasattr(boxes, '__len__') else 0
                sample_stats['total_boxes'] += num_boxes
                print(f"    📦 boxes_3d: {num_boxes} 个检测框")

            if 'scores_3d' in sample:
                sample_stats['has_scores_3d'] += 1
                scores = sample['scores_3d']
                if hasattr(scores, '__len__') and len(scores) > 0:
                    score_range = [float(np.min(scores)), float(np.max(scores))]
                    sample_stats['box_score_ranges'].append(score_range)
                    print(f"    🎯 scores_3d: 范围 [{score_range[0]:.4f}, {score_range[1]:.4f}]")
            if 'labels_3d' in sample:
                sample_stats['has_labels_3d'] += 1
                labels = sample['labels_3d']
                if hasattr(labels, '__len__') and len(labels) > 0:
                    unique_labels = np.unique(labels)
                    print(f"    🏷️ labels_3d: {len(unique_labels)} 个类别 {unique_labels}")

            # 检查BEV分割结果
            if 'masks_bev' in sample:
                sample_stats['has_masks_bev'] += 1
                masks = sample['masks_bev']
                if hasattr(masks, 'shape'):
                    shape = masks.shape
                    sample_stats['mask_shapes'].append(shape)
                    print(f"    🗺️ masks_bev: 形状 {shape}")
                    print(f"      分割类别数: {shape[0]}, BEV尺寸: {shape[1]}×{shape[2]}")

                    # 统计每个类别的像素占比
                    for c in range(min(3, shape[0])):  # 只显示前3个类别
                        if hasattr(masks, 'shape') and len(shape) >= 3:
                            pixel_count = np.sum(masks[c])
                            total_pixels = shape[1] * shape[2]
                            percentage = pixel_count / total_pixels * 100
                            print(f"        类别{c}: {pixel_count}像素 ({percentage:.2f}%)")
        else:
            print(f"    ⚠️ 意外的样本类型: {type(sample)}")

    # 汇总统计
    print(f"\n📈 统计汇总 (前{samples_to_analyze}个样本):")
    print(f"   📦 包含3D检测框: {sample_stats['has_boxes_3d']}/{samples_to_analyze}")
    print(f"   🎯 包含检测分数: {sample_stats['has_scores_3d']}/{samples_to_analyze}")
    print(f"   🏷️ 包含检测标签: {sample_stats['has_labels_3d']}/{samples_to_analyze}")
    print(f"   🗺️ 包含BEV分割: {sample_stats['has_masks_bev']}/{samples_to_analyze}")
    print(f"   📊 总检测框数: {sample_stats['total_boxes']}")

    if sample_stats['box_score_ranges']:
        all_scores = [r for ranges in sample_stats['box_score_ranges'] for r in ranges]
        print(f"   🎯 检测分数范围: [{min(all_scores):.4f}, {max(all_scores):.4f}]")
    if sample_stats['mask_shapes']:
        shapes = sample_stats['mask_shapes']
        print(f"   🔲 BEV分割形状: {shapes[0] if shapes else 'N/A'}")

def main():
    if len(sys.argv) != 2:
        print("用法: python INSPECT_FAST_RESULTS.py <results_file>")
        sys.exit(1)

    results_file = sys.argv[1]
    load_results_sample(results_file)

if __name__ == "__main__":
    main()