English version:

Generally introduction:

this project can be ued to recognize bikes' photos,which can utilize CUDA to accelerate the project. After using data to train the module, this module can judge a photo whether it is a bike or not.Simultaneously,the data can be replace by ohter kind of things because the data come from CIFAR,which includes various things(the detail can be seen in the data part below) .

training data

the training data come from CIFAR,which includes CIFAR-10 and CIFAR-100.

CIFAR-10:

Core Dataset Information​Number of Classes

10 mutually exclusive object categories (6,000 images per class, totaling 60,000 images).

​Image Dimensions

32×32-pixel RGB color images (extremely low resolution, optimized for fast training).

​Data Split

Training Set: 50,000 images (5,000 images per class). Test Set: 10,000 images (1,000 images per class).​

Data Source

Curated and labeled from the 80 Million Tiny Images dataset by Alex Krizhevsky.​2. The 10 Object Categories and Examples The CIFAR-10 dataset encompasses everyday objects, with the 10 categories explicitly defined as follows:

CLFAR-100

Number of Classes

100 fine-grained object categories (e.g., subtypes of vehicles, plants, or medical conditions), with ​60,000–100,000 images (scaling proportionally to CIFAR-10’s 60,000 images).

​Image Dimensions

High-resolution (e.g., 224×224 or 512×512 pixels), enabling detailed feature extraction for tasks like fine-grained classification and segmentation.

Data Split

Training Set: 80,000–90,000 images (800–900 images per class). Validation Set: 5,000 images (50 images per class). Test Set: 10,000–15,000 images (100–150 images per class).​Data Sources: Synthesized from ​real-world sensors (e.g., autonomous vehicles, medical imaging devices) and ​public datasets (e.g., ImageNet-21k, OpenImages). Annotated via ​semi-automated tools + ​human experts to ensure precision in fine-grained labels.

addtion

in this project,I downloaded the data in the dirction named "cifar-100-python".If you don't want to download it in your computer,you can also ues it,it just has a difference in the code.

coding part:

utilizing CUDA

import os 
# 检查GPU是否可用 to check if CUDA is available
device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')
print(f"使用设备: {device}")

on my device,I uesd CUDA 0.the uesr can change the number of the CUDA

traning data collection

if you download the data in your computer and place it in the same dirction with the code.

# 加载CIFAR-100数据集
train_set = torchvision.datasets.CIFAR100(root='./', train=True, download=False, transform=transform)
test_set = torchvision.datasets.CIFAR100(root='./', train=False, download=False, transform=transform)

'./' means use the data in the same dirction and 'download = False' means I have downloaded the data in my computer

creating a label for bike

#查找bicycle的索引 find the index of bicycle
bike_class_idx = train_set.classes.index('bicycle')
print(f"自行车（bicycle）的索引是: {bike_class_idx}")
print("查找成功")
# 创建自行车/非自行车二分类标签
def create_binary_labels(dataset,bike_class_idx):
    for i in range(len(dataset)):
        _, label = dataset[i]
        dataset.targets[i] = 1 if label == bike_class_idx else 0

if you ues this project for recognizing other things,you should find the right index and modify this part of codes.

function checking

the project also inludes drawing a gragh to show the function of the module

# 5. 绘制训练曲线
def plot_training(train_losses, val_losses, train_acc, val_acc):
    plt.figure(figsize=(12, 4))

    plt.subplot(1, 2, 1)
    plt.plot(train_losses, label='training losses')
    plt.plot(val_losses, label='val losses')
    plt.title('training and val losses')
    plt.xlabel('Epoch')
    plt.ylabel('Loss')
    plt.legend()

    plt.subplot(1, 2, 2)
    plt.plot(train_acc, label='training acc')
    plt.plot(val_acc, label='val acc')
    plt.title('training and val acc')
    plt.xlabel('Epoch')
    plt.ylabel('Accuracy (%)')
    plt.legend()

    plt.tight_layout()
    #plt.show()
    plt.savefig('training_plot.png')

a traning_plot.png will be saved in the dirction.

function test

# 7. 测试单张图像
def predict_bike(image_tensor):
    """预测单张图像是否为自行车"""
    model.eval()
    with torch.no_grad():
        image_tensor = image_tensor.unsqueeze(0).to(device)  # 添加批次维度
        output = model(image_tensor)
        probability = output.item()
        return "bicycle" if probability > 0.5 else "not bicycle", probability

this part of code selets a image at random from the traning module to let the module judge whether it is a bicycle or not

summarize

this project still has a lot of weakness,weclome everyone to give suggestions or pull request.Any useful issues will be carefully considered.If this project can help you,please give a star.