#!/usr/bin/env python3 """ Phase 4B RMT-PPAD 评估结果可视化脚本 解析评估结果文件并生成可视化输出 """ import argparse import pickle import os import numpy as np import matplotlib.pyplot as plt import seaborn as sns from pathlib import Path def parse_results(results_file): """解析MMDetection3D评估结果文件""" print(f"📂 读取结果文件: {results_file}") if not os.path.exists(results_file): print(f"❌ 文件不存在: {results_file}") return None with open(results_file, 'rb') as f: results = pickle.load(f) return results def visualize_bbox_metrics(results, save_path="bbox_metrics.png"): """可视化3D检测指标""" print("📊 可视化3D检测指标...") if 'bbox' not in results: print("⚠️ 没有找到bbox指标") return bbox_results = results['bbox'] # 解析不同类别的指标 classes = [] aps = [] ap_50s = [] ap_75s = [] for class_name, metrics in bbox_results.items(): if class_name.startswith('car') or class_name.startswith('truck') or \ class_name.startswith('bus') or class_name.startswith('pedestrian'): classes.append(class_name.split('_')[0]) # 去掉AP后缀 aps.append(metrics.get('ap', 0)) ap_50s.append(metrics.get('ap_0.5', 0)) ap_75s.append(metrics.get('ap_0.75', 0)) if not classes: print("⚠️ 没有找到有效的类别指标") return # 创建图表 fig, axes = plt.subplots(1, 3, figsize=(15, 5)) x = np.arange(len(classes)) width = 0.25 # AP@0.5:0.95 axes[0].bar(x - width, aps, width, label='AP@0.5:0.95', color='skyblue') axes[0].set_title('Average Precision (AP@0.5:0.95)') axes[0].set_xticks(x) axes[0].set_xticklabels(classes, rotation=45) axes[0].set_ylabel('AP') axes[0].grid(True, alpha=0.3) # AP@0.5 axes[1].bar(x, ap_50s, width, label='AP@0.5', color='lightgreen') axes[1].set_title('Average Precision (AP@0.5)') axes[1].set_xticks(x) axes[1].set_xticklabels(classes, rotation=45) axes[1].set_ylabel('AP') axes[1].grid(True, alpha=0.3) # AP@0.75 axes[2].bar(x + width, ap_75s, width, label='AP@0.75', color='salmon') axes[2].set_title('Average Precision (AP@0.75)') axes[2].set_xticks(x) axes[2].set_xticklabels(classes, rotation=45) axes[2].set_ylabel('AP') axes[2].grid(True, alpha=0.3) plt.tight_layout() plt.savefig(save_path, dpi=300, bbox_inches='tight') plt.close() print(f"✅ 3D检测指标图表已保存: {save_path}") def visualize_map_metrics(results, save_path="map_metrics.png"): """可视化BEV分割指标""" print("🗺️ 可视化BEV分割指标...") if 'map' not in results: print("⚠️ 没有找到map指标") return map_results = results['map'] # 解析分割类别 classes = [] ious = [] dices = [] for class_name, metrics in map_results.items(): if 'IoU' in metrics and 'Dice' in metrics: # 去掉类别名称中的数字后缀 clean_name = class_name.replace('_0', '').replace('_1', '').replace('_2', '') \ .replace('_3', '').replace('_4', '').replace('_5', '') if clean_name not in classes: classes.append(clean_name) ious.append(metrics['IoU']) dices.append(metrics['Dice']) if not classes: print("⚠️ 没有找到有效的分割指标") return # 创建图表 fig, ax = plt.subplots(figsize=(10, 6)) x = np.arange(len(classes)) width = 0.35 bars1 = ax.bar(x - width/2, ious, width, label='IoU', color='lightblue', alpha=0.8) bars2 = ax.bar(x + width/2, dices, width, label='Dice', color='lightcoral', alpha=0.8) ax.set_xlabel('类别') ax.set_ylabel('分数') ax.set_title('BEV分割性能指标') ax.set_xticks(x) ax.set_xticklabels(classes, rotation=45, ha='right') ax.legend() ax.grid(True, alpha=0.3) # 添加数值标签 for bar, value in zip(bars1, ious): height = bar.get_height() ax.text(bar.get_x() + bar.get_width()/2., height + 0.01, f'{value:.3f}', ha='center', va='bottom', fontsize=8) for bar, value in zip(bars2, dices): height = bar.get_height() ax.text(bar.get_x() + bar.get_width()/2., height + 0.01, f'{value:.3f}', ha='center', va='bottom', fontsize=8) plt.tight_layout() plt.savefig(save_path, dpi=300, bbox_inches='tight') plt.close() print(f"✅ BEV分割指标图表已保存: {save_path}") def print_summary(results, log_file=None): """打印评估结果摘要""" print("\n" + "="*60) print("📈 Phase 4B RMT-PPAD 评估结果摘要") print("="*60) # 3D检测指标 if 'bbox' in results: print("\n🚗 3D检测性能:") bbox_results = results['bbox'] # 计算NDS (如果有的话) if 'NDS' in bbox_results: print(".4f") # 各类别AP for class_name, metrics in bbox_results.items(): if isinstance(metrics, dict) and 'ap' in metrics: print(f" {class_name}: AP@0.5:0.95 = {metrics.get('ap', 0):.4f}, " f"AP@0.5 = {metrics.get('ap_0.5', 0):.4f}") # BEV分割指标 if 'map' in results: print("\n🗺️ BEV分割性能:") map_results = results['map'] total_iou = 0 total_dice = 0 count = 0 for class_name, metrics in map_results.items(): if isinstance(metrics, dict) and 'IoU' in metrics: iou = metrics['IoU'] dice = metrics.get('Dice', 0) print(f" {class_name}: IoU = {iou:.4f}, Dice = {dice:.4f}") total_iou += iou total_dice += dice count += 1 if count > 0: print(f" 平均IoU: {total_iou/count:.4f}, 平均Dice: {total_dice/count:.4f}") print("\n" + "="*60) # 保存到日志文件 if log_file: with open(log_file, 'w') as f: f.write("Phase 4B RMT-PPAD 评估结果摘要\n\n") f.write("生成时间: " + str(np.datetime64('now')) + "\n\n") if 'bbox' in results: f.write("3D检测性能:\n") for class_name, metrics in results['bbox'].items(): if isinstance(metrics, dict) and 'ap' in metrics: f.write(f" {class_name}: AP@0.5:0.95 = {metrics.get('ap', 0):.4f}\n") if 'map' in results: f.write("\nBEV分割性能:\n") for class_name, metrics in results['map'].items(): if isinstance(metrics, dict) and 'IoU' in metrics: f.write(f" {class_name}: IoU = {metrics['IoU']:.4f}\n") print(f"📝 详细结果已保存到: {log_file}") def create_visualization_script(results_file, output_dir="visualization"): """创建可视化脚本的主函数""" os.makedirs(output_dir, exist_ok=True) # 解析结果 results = parse_results(results_file) if results is None: return # 生成可视化 bbox_plot = os.path.join(output_dir, "bbox_metrics.png") map_plot = os.path.join(output_dir, "map_metrics.png") summary_file = os.path.join(output_dir, "evaluation_summary.txt") visualize_bbox_metrics(results, bbox_plot) visualize_map_metrics(results, map_plot) print_summary(results, summary_file) print(f"\n🎉 可视化完成！输出目录: {output_dir}") print(f" 📊 3D检测图表: {bbox_plot}") print(f" 🗺️ 分割图表: {map_plot}") print(f" 📝 结果摘要: {summary_file}") def main(): parser = argparse.ArgumentParser(description="Phase 4B RMT-PPAD 评估结果可视化工具") parser.add_argument("results_file", help="评估结果文件路径 (.pkl)") parser.add_argument("--out-dir", default="visualization", help="输出目录 (默认: visualization)") args = parser.parse_args() print("🚀 Phase 4B RMT-PPAD 评估结果可视化工具") print(f"📁 结果文件: {args.results_file}") print(f"📂 输出目录: {args.out_dir}") print() create_visualization_script(args.results_file, args.out_dir) if __name__ == "__main__": main()