Files

300 lines
9.1 KiB
Python

"""
PC端演示脚本 - 适合录屏展示
功能:
1. 加载训练好的模型
2. 逐张显示测试图像
3. 实时显示预测结果(类别、置信度、推理时间)
4. 适合录屏演示给老师
"""
import os
import time
import torch
import torch.nn as nn
from torchvision import transforms
from PIL import Image
import matplotlib.pyplot as plt
import matplotlib
import numpy as np
# 设置中文字体
matplotlib.rcParams['font.sans-serif'] = ['SimHei'] # 黑体
matplotlib.rcParams['axes.unicode_minus'] = False
from model import create_model
def load_model(model_path, num_classes=10, device='cpu'):
"""加载训练好的模型"""
print(f"正在加载模型: {model_path}")
model = create_model(num_classes=num_classes, pretrained=False)
checkpoint = torch.load(model_path, map_location=device)
if 'model_state_dict' in checkpoint:
model.load_state_dict(checkpoint['model_state_dict'])
else:
model.load_state_dict(checkpoint)
model = model.to(device)
model.eval()
print("模型加载完成\n")
return model
def predict_image(model, image_path, transform, device):
"""对单张图像进行预测"""
# 加载图像
image = Image.open(image_path).convert('RGB')
image_tensor = transform(image).unsqueeze(0).to(device)
# 推理
with torch.no_grad():
start_time = time.time()
output = model(image_tensor)
inference_time = (time.time() - start_time) * 1000 # ms
probabilities = torch.softmax(output, dim=1)
pred_class = torch.argmax(probabilities, dim=1).item()
confidence = probabilities[0][pred_class].item()
return image, pred_class, confidence, inference_time, probabilities[0].cpu().numpy()
def display_prediction(image, pred_class, confidence, inference_time,
true_class, probabilities, class_names):
"""显示预测结果"""
fig = plt.figure(figsize=(14, 6))
# 左侧:显示图像
ax1 = plt.subplot(1, 2, 1)
ax1.imshow(image)
ax1.axis('off')
# 标题显示预测结果
is_correct = (pred_class == true_class)
color = 'green' if is_correct else 'red'
result_text = '正确' if is_correct else '错误'
title = f'真实类别: {class_names[true_class]}\n'
title += f'预测类别: {class_names[pred_class]} ({result_text})\n'
title += f'置信度: {confidence*100:.2f}%\n'
title += f'推理时间: {inference_time:.2f} ms'
ax1.set_title(title, fontsize=14, color=color, fontweight='bold')
# 右侧:显示概率分布
ax2 = plt.subplot(1, 2, 2)
bars = ax2.barh(class_names, probabilities * 100)
# 高亮预测类别
bars[pred_class].set_color('red')
bars[true_class].set_edgecolor('green')
bars[true_class].set_linewidth(3)
ax2.set_xlabel('置信度 (%)', fontsize=12)
ax2.set_title('各类别预测概率', fontsize=14, fontweight='bold')
ax2.set_xlim(0, 100)
# 添加数值标签
for i, (bar, prob) in enumerate(zip(bars, probabilities)):
if prob > 0.01: # 只显示>1%的概率
ax2.text(prob * 100 + 1, i, f'{prob*100:.1f}%',
va='center', fontsize=10)
plt.tight_layout()
plt.pause(0.1) # 短暂暂停以更新显示
return fig
def demo_single_mode(model, data_dir, transform, device, class_names):
"""单张演示模式 - 逐张显示"""
print("\n" + "="*60)
print("单张演示模式 - 按任意键继续,按 'q' 退出")
print("="*60 + "\n")
# 获取所有测试图像
image_paths = []
true_labels = []
for class_idx, class_name in enumerate(class_names):
class_dir = os.path.join(data_dir, class_name)
if os.path.exists(class_dir):
for img_name in os.listdir(class_dir):
if img_name.endswith(('.jpg', '.png', '.jpeg')):
image_paths.append(os.path.join(class_dir, img_name))
true_labels.append(class_idx)
print(f"找到 {len(image_paths)} 张测试图像\n")
correct = 0
total = 0
plt.ion() # 开启交互模式
for idx, (img_path, true_label) in enumerate(zip(image_paths, true_labels)):
print(f"\n[{idx+1}/{len(image_paths)}] 正在处理: {os.path.basename(img_path)}")
# 预测
image, pred_class, confidence, inf_time, probs = predict_image(
model, img_path, transform, device
)
# 显示结果
fig = display_prediction(
image, pred_class, confidence, inf_time,
true_label, probs, class_names
)
# 统计
if pred_class == true_label:
correct += 1
total += 1
# 打印结果
result = "正确" if pred_class == true_label else "错误"
print(f" 真实: {class_names[true_label]}")
print(f" 预测: {class_names[pred_class]} ({result})")
print(f" 置信度: {confidence*100:.2f}%")
print(f" 推理时间: {inf_time:.2f} ms")
print(f" 当前准确率: {correct}/{total} = {correct/total*100:.2f}%")
# 等待用户输入
print("\n按 Enter 继续,输入 'q' 退出...")
user_input = input()
plt.close(fig)
if user_input.lower() == 'q':
break
plt.ioff()
# 最终统计
print("\n" + "="*60)
print("演示完成!")
print(f"总体准确率: {correct}/{total} = {correct/total*100:.2f}%")
print("="*60)
def demo_auto_mode(model, data_dir, transform, device, class_names, delay=2.0):
"""自动演示模式 - 自动播放,适合录屏"""
print("\n" + "="*60)
print(f"自动演示模式 - 每张图像停留 {delay} 秒")
print("="*60 + "\n")
# 获取所有测试图像
image_paths = []
true_labels = []
for class_idx, class_name in enumerate(class_names):
class_dir = os.path.join(data_dir, class_name)
if os.path.exists(class_dir):
for img_name in sorted(os.listdir(class_dir)):
if img_name.endswith(('.jpg', '.png', '.jpeg')):
image_paths.append(os.path.join(class_dir, img_name))
true_labels.append(class_idx)
print(f"找到 {len(image_paths)} 张测试图像\n")
print("开始演示...\n")
correct = 0
total = 0
total_time = 0
plt.ion() # 开启交互模式
for idx, (img_path, true_label) in enumerate(zip(image_paths, true_labels)):
print(f"[{idx+1}/{len(image_paths)}] {os.path.basename(img_path)}", end=" ")
# 预测
image, pred_class, confidence, inf_time, probs = predict_image(
model, img_path, transform, device
)
# 显示结果
fig = display_prediction(
image, pred_class, confidence, inf_time,
true_label, probs, class_names
)
# 统计
if pred_class == true_label:
correct += 1
print("[正确]", end=" ")
else:
print("[错误]", end=" ")
total += 1
total_time += inf_time
print(f"准确率: {correct/total*100:.1f}%")
# 等待
plt.pause(delay)
plt.close(fig)
plt.ioff()
# 最终统计
print("\n" + "="*60)
print("演示完成!")
print(f"总体准确率: {correct}/{total} = {correct/total*100:.2f}%")
print(f"平均推理时间: {total_time/total:.2f} ms")
print(f"吞吐量: {1000/(total_time/total):.2f} FPS")
print("="*60)
def main():
import argparse
parser = argparse.ArgumentParser(description='PC端演示脚本')
parser.add_argument('--model_path', type=str,
default='data/models/best_model.pth',
help='模型路径')
parser.add_argument('--data_dir', type=str,
default='data/processed/val',
help='测试数据目录')
parser.add_argument('--mode', type=str, default='auto',
choices=['single', 'auto'],
help='演示模式: single(单张) 或 auto(自动)')
parser.add_argument('--delay', type=float, default=2.0,
help='自动模式下每张图像停留时间(秒)')
parser.add_argument('--device', type=str, default='cpu',
help='设备 (cpu/cuda)')
args = parser.parse_args()
# 设置设备
device = torch.device(args.device if torch.cuda.is_available() else 'cpu')
# 数据预处理
transform = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
])
# 类别名称
class_names = [f'class_{i}' for i in range(10)]
print("="*60)
print("PC端光图像识别演示系统")
print("="*60)
print(f"模型: MobileNetV2")
print(f"设备: {device}")
print(f"模式: {args.mode}")
# 加载模型
model = load_model(args.model_path, num_classes=10, device=device)
# 运行演示
if args.mode == 'single':
demo_single_mode(model, args.data_dir, transform, device, class_names)
else:
demo_auto_mode(model, args.data_dir, transform, device, class_names, args.delay)
if __name__ == '__main__':
main()