数据集简介：

训练集共有2160张猫的图片, 分为12类. train_list.txt是其标注文件

测试集共有240张猫的图片. 不含标注信息.

训练集图像（部分）

验证集图像（部分）

标签

部分代码：

# 定义训练数据集
class TrainData(Dataset):
    def __init__(self):
        super().__init__()
        self.color_jitter = T.ColorJitter(brightness=0.05, contrast=0.05, saturation=0.05, hue=0.05)
        self.normalize = T.Normalize(mean=0, std=1)
        self.random_crop = T.RandomCrop(224, pad_if_needed=True)
    
    def __getitem__(self, index):
        # 读取图片
        image_path = train_paths[index]

        image = np.array(Image.open(image_path))    # H, W, C
        try:
            image = image.transpose([2, 0, 1])[:3]  # C, H, W
        except:
            image = np.array([image, image, image]) # C, H, W
        
        # 图像增广
        features = self.color_jitter(image.transpose([1, 2, 0]))
        features = self.random_crop(features)
        features = self.normalize(features.transpose([2, 0, 1])).astype(np.float32)

        # 读取标签
        labels = train_labels[index]

        return features, labels
    
    def __len__(self):
        return len(train_paths)

    
# 定义验证数据集
class ValidData(Dataset):
    def __init__(self):
        super().__init__()
        self.normalize = T.Normalize(mean=0, std=1)
    
    def __getitem__(self, index):
        # 读取图片
        image_path = valid_paths[index]

        image = np.array(Image.open(image_path))    # H, W, C
        try:
            image = image.transpose([2, 0, 1])[:3]  # C, H, W
        except:
            image = np.array([image, image, image]) # C, H, W
        
        # 图像变换
        features = cv2.resize(image.transpose([1, 2, 0]), (256, 256)).transpose([2, 0, 1]).astype(np.float32)
        features = self.normalize(features)

        # 读取标签
        labels = valid_labels[index]

        return features, labels
    
    def __len__(self):
        return len(valid_paths)

# 调用resnet50模型
paddle.vision.set_image_backend('cv2')
model = paddle.vision.models.resnet50(pretrained=True, num_classes=12)

# 定义数据迭代器
train_dataloader = DataLoader(train_data, batch_size=256, shuffle=True, drop_last=False)

# 定义优化器
opt = paddle.optimizer.Adam(learning_rate=1e-4, parameters=model.parameters(), weight_decay=paddle.regularizer.L2Decay(1e-4))

# 定义损失函数
loss_fn = paddle.nn.CrossEntropyLoss()

# 设置gpu环境
paddle.set_device('gpu:0')

# 整体训练流程
for epoch_id in range(15):
    model.train()
    for batch_id, data in enumerate(train_dataloader()):
        # 读取数据
        features, labels = data
        features = paddle.to_tensor(features)
        labels = paddle.to_tensor(labels)

        # 前向传播
        predicts = model(features)

        # 损失计算
        loss = loss_fn(predicts, labels)

        # 反向传播
        avg_loss = paddle.mean(loss)
        avg_loss.backward()

        # 更新
        opt.step()

        # 清零梯度
        opt.clear_grad()

        # 打印损失
        if batch_id % 2 == 0:
            print('epoch_id:{}, batch_id:{}, loss:{}'.format(epoch_id, batch_id, avg_loss.numpy()))
    model.eval()
    print('开始评估')
    i = 0
    acc = 0
    for image, label in valid_data:
        image = paddle.to_tensor([image])

        pre = list(np.array(model(image)[0]))
        max_item = max(pre)
        pre = pre.index(max_item)

        i += 1
        if pre == label:
            acc += 1
        if i % 10 == 0:
            print('精度：', acc / i)
    
    paddle.save(model.state_dict(), 'acc{}.model'.format(acc / i))

# 进行预测和提交
# 首先拿到预测文件的路径列表

def listdir(path, list_name):
    for file in os.listdir(path):
        file_path = os.path.join(path, file)
        if os.path.isdir(file_path):
            listdir(file_path, list_name)
        else:
            list_name.append(file_path)
test_path = []
listdir('cat_12_test', test_path)

# 加载训练好的模型
pre_model = paddle.vision.models.resnet50(pretrained=True, num_classes=12)
pre_model.set_state_dict(paddle.load('acc0.9285714285714286.model'))
pre_model.eval()

pre_classes = []
normalize = T.Normalize(mean=0, std=1)
# 生成预测结果
for path in test_path:
    image_path = path

    image = np.array(Image.open(image_path))    # H, W, C
    try:
        image = image.transpose([2, 0, 1])[:3]  # C, H, W
    except:
        image = np.array([image, image, image]) # C, H, W
    
    # 图像变换
    features = cv2.resize(image.transpose([1, 2, 0]), (256, 256)).transpose([2, 0, 1]).astype(np.float32)
    features = normalize(features)

    features = paddle.to_tensor([features])
    pre = list(np.array(pre_model(features)[0]))
    # print(pre)
    max_item = max(pre)
    pre = pre.index(max_item)
    print("图片：", path, "预测结果：", pre)
    pre_classes.append(pre)

print(pre_classes)

数据集链接：猫类识别-12种猫分类数据集（2400张）