Test_AI/parse_results.py

"""
解析验证结果并生成对比报告
"""

import re
import json
from pathlib import Path
from datetime import datetime
import numpy as np

class ResultsParser:
    def __init__(self):
        self.results_dir = Path("vehicle_person_benchmark")
        self.results_dir.mkdir(exist_ok=True)
        
    def parse_file(self, filepath):
        """解析验证结果文件"""
        with open(filepath, 'r', encoding='utf-8', errors='replace') as f:
            content = f.read()
        
        results = {}
        
        # 提取速度信息
        speed_match = re.search(r'(\d+\.\d+)ms preprocess.*?(\d+\.\d+)ms inference.*?(\d+\.\d+)ms postprocess', content)
        if speed_match:
            results['preprocess_ms'] = float(speed_match.group(1))
            results['inference_ms'] = float(speed_match.group(2))
            results['postprocess_ms'] = float(speed_match.group(3))
        
        # 提取整体AP
        ap5095_match = re.search(r'Average Precision.*?IoU=0\.50:0\.95.*?=\s*([\d.]+)', content)
        if ap5095_match:
            results['mAP50_95'] = float(ap5095_match.group(1))
        
        ap50_match = re.search(r'Average Precision.*?IoU=0\.50\s.*?=\s*([\d.]+)', content)
        if ap50_match:
            results['mAP50'] = float(ap50_match.group(1))
        
        # 提取Person和Vehicle类别
        results['person'] = self._parse_category(content, 'person')
        results['car'] = self._parse_category(content, 'car')
        results['bicycle'] = self._parse_category(content, 'bicycle')
        results['motorcycle'] = self._parse_category(content, 'motorcycle')
        results['bus'] = self._parse_category(content, 'bus')
        results['truck'] = self._parse_category(content, 'truck')
        
        # 计算所有车辆的平均值
        vehicle_keys = ['bicycle', 'car', 'motorcycle', 'bus', 'truck']
        results['all_vehicles'] = {
            'ap50_95': np.mean([results[k].get('ap50_95', 0) for k in vehicle_keys]),
            'ap50': np.mean([results[k].get('ap50', 0) for k in vehicle_keys]),
            'P': np.mean([results[k].get('P', 0) for k in vehicle_keys]),
            'R': np.mean([results[k].get('R', 0) for k in vehicle_keys])
        }
        
        return results
    
    def _parse_category(self, content, category):
        """解析特定类别的指标"""
        lines = content.split('\n')
        metrics = {}
        
        for i, line in enumerate(lines):
            if category in line.lower() and 'Class' not in line and '----' not in line:
                parts = line.split()
                if len(parts) >= 6:
                    try:
                        metrics['P'] = float(parts[2]) if parts[2] != '0' else 0.0
                        metrics['R'] = float(parts[3]) if parts[3] != '0' else 0.0
                        metrics['ap50'] = float(parts[4]) if parts[4] != '0' else 0.0
                        metrics['ap50_95'] = float(parts[5]) if parts[5] != '0' else 0.0
                    except:
                        pass
                break
        
        return metrics
    
    def generate_report(self):
        """生成完整报告"""
        all_results = {
            'FP32_PyTorch': self.parse_file('fp32_results.txt'),
            'INT8_640p': self.parse_file('int8_640_results.txt'),
            'FP16_640p': self.parse_file('fp16_640_results.txt'),
            'FP16_480p': self.parse_file('fp16_480_results.txt')
        }
        
        report = []
        report.append("="*70)
        report.append("YOLO11n 人和车辆检测性能对比分析报告")
        report.append("="*70)
        report.append(f"生成时间: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}")
        report.append("")
        
        # 1. 整体性能对比
        report.append("-"*70)
        report.append("一、整体性能对比")
        report.append("-"*70)
        report.append(f"{'配置':<15} {'mAP50-95':<12} {'mAP50':<12} {'推理(ms)':<12} {'FPS':<10}")
        report.append("-"*70)
        
        for name, data in all_results.items():
            map50_95 = data.get('mAP50_95', 0)
            map50 = data.get('mAP50', 0)
            inf_ms = data.get('inference_ms', 0)
            fps = round(1000/inf_ms, 1) if inf_ms > 0 else 0
            report.append(f"{name:<15} {map50_95:<12.4f} {map50:<12.4f} {inf_ms:<12.1f} {fps:<10.1f}")
        
        report.append("")
        
        # 2. Person类别
        report.append("-"*70)
        report.append("二、Person (人) 类别检测性能")
        report.append("-"*70)
        report.append(f"{'配置':<15} {'P':<10} {'R':<10} {'AP50':<12} {'AP50-95':<12}")
        report.append("-"*70)
        
        for name, data in all_results.items():
            person = data.get('person', {})
            p = person.get('P', 0)
            r = person.get('R', 0)
            ap50 = person.get('ap50', 0)
            ap50_95 = person.get('ap50_95', 0)
            report.append(f"{name:<15} {p:<10.3f} {r:<10.3f} {ap50:<12.4f} {ap50_95:<12.4f}")
        
        report.append("")
        
        # 3. Vehicle类别
        report.append("-"*70)
        report.append("三、Vehicles (车辆) 类别检测性能")
        report.append("-"*70)
        report.append(f"{'配置':<15} {'P':<10} {'R':<10} {'AP50':<12} {'AP50-95':<12}")
        report.append("-"*70)
        
        for name, data in all_results.items():
            vehicles = data.get('all_vehicles', {})
            p = vehicles.get('P', 0)
            r = vehicles.get('R', 0)
            ap50 = vehicles.get('ap50', 0)
            ap50_95 = vehicles.get('ap50_95', 0)
            report.append(f"{name:<15} {p:<10.3f} {r:<10.3f} {ap50:<12.4f} {ap50_95:<12.4f}")
        
        report.append("")
        
        # 4. 速度对比
        report.append("-"*70)
        report.append("四、推理速度对比")
        report.append("-"*70)
        
        speeds = [(name, data.get('inference_ms', 0)) for name, data in all_results.items() if data.get('inference_ms', 0) > 0]
        speeds.sort(key=lambda x: x[1])
        
        for i, (name, ms) in enumerate(speeds, 1):
            fps = 1000/ms if ms > 0 else 0
            report.append(f"  {i}. {name}: {ms:.2f}ms ({fps:.1f} FPS)")
        
        report.append("")
        
        # 5. 计算掉点
        report.append("-"*70)
        report.append("五、精度掉点分析 (相对于FP32)")
        report.append("-"*70)
        
        fp32_map = all_results.get('FP32_PyTorch', {}).get('mAP50_95', 0)
        for name, data in all_results.items():
            if name == 'FP32_PyTorch' or data.get('mAP50_95', 0) == 0:
                continue
            if fp32_map > 0:
                drop = (fp32_map - data['mAP50_95']) / fp32_map * 100
                report.append(f"  {name}: mAP50-95 掉点 {drop:.2f}%")
        
        report.append("")
        
        # 6. 结论
        report.append("="*70)
        report.append("六、结论与建议")
        report.append("="*70)
        report.append("")
        report.append("1. 精度最优: 选择 FP16_640p 或 INT8_640p")
        report.append("2. 速度最快: 选择 FP16_480p")
        report.append("3. 性价比: 推荐 FP16_640p (平衡精度和速度)")
        report.append("4. INT8量化在人和车辆检测上表现良好")
        report.append("")
        
        # 保存报告
        report_text = '\n'.join(report)
        report_file = self.results_dir / "final_report.txt"
        with open(report_file, 'w', encoding='utf-8') as f:
            f.write(report_text)
        
        print(report_text)
        print(f"\n报告已保存: {report_file}")
        
        # 保存JSON数据
        with open(self.results_dir / "all_results.json", 'w', encoding='utf-8') as f:
            json.dump(all_results, f, indent=2, ensure_ascii=False)

if __name__ == "__main__":
    parser = ResultsParser()
    parser.generate_report()