diff --git a/cv/README.md b/cv/README.md
index d8edae0..223bb8c 100644
--- a/cv/README.md
+++ b/cv/README.md
@@ -9,6 +9,7 @@ This directory contains ready-to-use Computer Vision application notebooks built
 | 1   | [ResNet](./resnet/) | Includes notebooks for ResNet finetuning on tasks such as chinese herbal classification |
 | 2   | [U-Net](./unet/) | Includes notebooks for U-Net training on tasks such as segmentation |
 | 3   | [SAM](./sam/) | Includes notebooks for using SAM to inference |
+| 4   | [Swin Transformer](./swin/) | Includes notebooks for Swin Transformer finetuning on tasks such as image classification |
 
 ## Contributing New CV Applications
 
diff --git a/cv/swin/finetune_swin_tiny_image_classification.ipynb b/cv/swin/finetune_swin_tiny_image_classification.ipynb
new file mode 100644
index 0000000..6cc65ff
--- /dev/null
+++ b/cv/swin/finetune_swin_tiny_image_classification.ipynb
@@ -0,0 +1,462 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "id": "0",
+   "metadata": {},
+   "source": [
+    "# 基于 Swin-Tiny 的图像分类微调\n",
+    "\n",
+    "## 1. 案例简介\n",
+    "\n",
+    "本案例展示了如何使用 MindSpore 和 MindSpore NLP 库，在自定义数据集（EuroSAT）上微调预训练的 Swin-Tiny 视觉 Transformer 模型。\n",
+    "\n",
+    "**Swin Transformer** 是一种分层的 Vision Transformer，其通过移动窗口计算自注意力（Shifted Windows），在图像分类、目标检测等任务上取得了优异的性能。Swin-Tiny 是该系列中轻量级的版本，适合在算力有限或需要快速迭代的场景下使用。\n",
+    "\n",
+    "### 任务流程\n",
+    "1. **环境准备**：安装并导入必要的库。\n",
+    "2. **数据加载**：加载 EuroSAT 卫星图像数据集。\n",
+    "3. **数据预处理**：使用 MindSpore Dataset 对图像进行 Resize、归一化等操作。\n",
+    "4. **模型构建**：加载预训练的 `swin-tiny-patch4-window7-224` 模型，并修改分类头。\n",
+    "5. **模型微调**：配置训练参数，使用 `Trainer` 接口进行训练。\n",
+    "6. **推理演示**：加载微调后的模型对新图像进行预测。\n",
+    "\n",
+    "### 运行环境\n",
+    "\n",
+    "| Python | MindSpore | MindSpore NLP |\n",
+    "| :----- | :-------- | :------------ |\n",
+    "| 3.10   | 2.7.0     | 0.5.1  |"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "1",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "#若在https://internstudio-ascend.intern-ai.org.cn/console/instance进行开发时，使用notebook会出现无法正常使用NPU，可进行以下步骤：\n",
+    "# 进入开发机的命令窗口\n",
+    "# 1. 激活你的环境\n",
+    "# conda activate mind_py310\n",
+    "# pip install ipykernel\n",
+    "# python -m ipykernel install --user --name=mind_py310 --display-name=\"Python (mind_py310)\"\n",
+    "# 2. 加载系统基础驱动配置\n",
+    "# source /usr/local/Ascend/ascend-toolkit/set_env.sh\n",
+    "# 3.【核心步骤】手动补全深层驱动路径 (修复 libascend_hal.so 报错)\n",
+    "# export LD_LIBRARY_PATH=/usr/local/Ascend/driver/lib64/driver:/usr/local/Ascend/driver/lib64/common:/usr/local/Ascend/driver/lib64:$LD_LIBRARY_PATH\n",
+    "# 4. 启动 Jupyter Lab\n",
+    "#jupyter lab --allow-root"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "2",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# 安装依赖\n",
+    "# !pip install mindnlp==0.5.1\n",
+    "# !pip install scikit-learn"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "3",
+   "metadata": {},
+   "source": [
+    "## 2. 环境配置\n",
+    "首先，我们需要应用一个补丁来修复 mindtorch 的版本兼容性问题，并配置 MindSpore 的运行环境。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "4",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# --------兼容性补丁---------\n",
+    "import mindtorch.autograd.function\n",
+    "if not hasattr(mindtorch.autograd.function, 'FunctionCtx'):\n",
+    "    class FunctionCtx:\n",
+    "        def __init__(self):\n",
+    "            self.saved_tensors = ()\n",
+    "        def save_for_backward(self, *tensors):\n",
+    "            self.saved_tensors = tensors\n",
+    "    mindtorch.autograd.function.FunctionCtx = FunctionCtx\n",
+    "    print(\"已应用 mindtorch 兼容性修复\")\n",
+    "\n",
+    "# ----基础配置与环境 ----\n",
+    "import mindnlp\n",
+    "import mindspore\n",
+    "import os\n",
+    "\n",
+    "# 设置 Token\n",
+    "os.environ[\"HF_TOKEN\"] = \"hf_**\" \n",
+    "\n",
+    "# 清理离线环境变量，确保联网\n",
+    "if 'HF_DATASETS_OFFLINE' in os.environ: del os.environ['HF_DATASETS_OFFLINE']\n",
+    "if 'TRANSFORMERS_OFFLINE' in os.environ: del os.environ['TRANSFORMERS_OFFLINE']\n",
+    "\n",
+    "# 设置随机种子\n",
+    "mindspore.set_seed(42)\n",
+    "print(\"环境配置完成\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "5",
+   "metadata": {},
+   "source": [
+    "## 3. 数据集加载与预处理\n",
+    "\n",
+    "我们将使用 **EuroSAT** 数据集，这是一个基于 Sentinel-2 卫星图像的土地利用和土地覆盖分类数据集，包含 10 个类别（如森林、河流、高速公路等）。由于 EuroSAT 只有 'train' 分割，我们需要将其划分为训练集和验证集。同时，我们需要构建标签 ID 到标签名称的映射字典。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "6",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from mindnlp.dataset import load_dataset\n",
+    "# 加载 EuroSAT 数据\n",
+    "dataset = load_dataset(\"jonathan-roberts1/EuroSAT\", split=\"train\")\n",
+    "\n",
+    "# 定义标签映射\n",
+    "labels = [\"AnnualCrop\", \"Forest\", \"HerbaceousVegetation\", \"Highway\", \"Industrial\", \n",
+    "          \"Pasture\", \"PermanentCrop\", \"Residential\", \"River\", \"SeaLake\"]\n",
+    "\n",
+    "label2id = {label: i for i, label in enumerate(labels)}\n",
+    "id2label = {i: label for i, label in enumerate(labels)}\n",
+    "\n",
+    "# 切分数据集\n",
+    "train_dataset, eval_dataset = dataset.split([0.9, 0.1])"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "7",
+   "metadata": {},
+   "source": [
+    "### 图像预处理\n",
+    "\n",
+    "为了适配 Swin Transformer 模型的输入要求并提升模型泛化能力，我们需要对数据进行特定的预处理。\n",
+    "\n",
+    "我们使用 `AutoImageProcessor` 自动加载模型对应的配置（如均值、方差、输入尺寸），并定义了两套不同的处理管道：\n",
+    "\n",
+    "* **训练集 (Training)**: 引入随机增强策略（随机裁剪、水平翻转），增加数据多样性，防止过拟合。\n",
+    "* **验证集 (Evaluation)**: 仅进行确定性的调整大小和中心裁剪，确保评估结果的一致性。\n",
+    "* **通用处理**: 包括像素归一化 (Normalize) 和维度转换 (HWC 转 CHW)。\n"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "8",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "import mindspore.dataset.vision as vision\n",
+    "import mindspore.dataset.transforms as transforms\n",
+    "from mindnlp.transformers import AutoImageProcessor\n",
+    "\n",
+    "model_checkpoint = \"microsoft/swin-tiny-patch4-window7-224\"\n",
+    "image_processor = AutoImageProcessor.from_pretrained(model_checkpoint)\n",
+    "\n",
+    "\n",
+    "# 自动获取模型所需的输入尺寸 (适配不同模型配置)\n",
+    "if \"shortest_edge\" in image_processor.size:\n",
+    "    size = image_processor.size[\"shortest_edge\"]\n",
+    "    crop_size = (size, size)\n",
+    "elif \"height\" in image_processor.size:\n",
+    "    size = (image_processor.size[\"height\"], image_processor.size[\"width\"])\n",
+    "    crop_size = size\n",
+    "    \n",
+    "print(f\"Image Processor Size Configuration: Resize={size}, Crop={crop_size}\")\n",
+    "\n",
+    "# 训练集增强策略\n",
+    "def get_train_transforms():\n",
+    "    \"\"\"\n",
+    "    1. RandomResizedCrop: 随机裁剪并缩放 (让模型学会看局部)\n",
+    "    2. RandomHorizontalFlip: 随机水平翻转 (让模型学会左右不变性)\n",
+    "    3. Rescale + Normalize: 归一化 (对应 ToTensor + Normalize)\n",
+    "    \"\"\"\n",
+    "    return transforms.Compose([\n",
+    "        vision.RandomResizedCrop(crop_size, scale=(0.08, 1.0), ratio=(0.75, 1.333)),\n",
+    "        vision.RandomHorizontalFlip(prob=0.5),\n",
+    "        vision.Rescale(1.0 / 255.0, 0.0), \n",
+    "        vision.Normalize(mean=image_processor.image_mean, std=image_processor.image_std),\n",
+    "        vision.HWC2CHW()\n",
+    "    ])\n",
+    "    \n",
+    "# 验证集增强策略\n",
+    "def get_eval_transforms():\n",
+    "    \"\"\"\n",
+    "    1. Resize: 调整大小\n",
+    "    2. CenterCrop: 中心裁剪 (保证每次评估输入一致)\n",
+    "    3. Rescale + Normalize: 归一化\n",
+    "    \"\"\"\n",
+    "    return transforms.Compose([\n",
+    "        vision.Resize(size),\n",
+    "        vision.CenterCrop(crop_size),\n",
+    "        vision.Rescale(1.0 / 255.0, 0.0),\n",
+    "        vision.Normalize(mean=image_processor.image_mean, std=image_processor.image_std),\n",
+    "        vision.HWC2CHW()\n",
+    "    ])\n",
+    "\n",
+    "# 定义两个处理函数\n",
+    "def train_transform_func(image, label):\n",
+    "    trans = get_train_transforms()\n",
+    "    return trans(image), label\n",
+    "\n",
+    "def eval_transform_func(image, label):\n",
+    "    trans = get_eval_transforms()\n",
+    "    return trans(image), label\n",
+    "\n",
+    "# 训练集应用增强策略\n",
+    "train_dataset = train_dataset.map(\n",
+    "    operations=train_transform_func, \n",
+    "    input_columns=[\"image\", \"label\"], \n",
+    "    output_columns=[\"pixel_values\", \"labels\"]\n",
+    ")\n",
+    "\n",
+    "# 验证集应用确定性策略\n",
+    "eval_dataset = eval_dataset.map(\n",
+    "    operations=eval_transform_func, \n",
+    "    input_columns=[\"image\", \"label\"], \n",
+    "    output_columns=[\"pixel_values\", \"labels\"]\n",
+    ")    "
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "9",
+   "metadata": {},
+   "source": [
+    "### 数据集内存化适配\n",
+    "\n",
+    "为了解决 `Trainer` 在处理流式数据集（IterableDataset）时可能出现的兼容性问题（如 `not subscriptable` 或 `datapipes` 缺失报错），同时充分利用服务器的大内存优势（240GB RAM），我们将处理后的数据集一次性加载到内存列表中。\n",
+    "\n",
+    "**核心优势：**\n",
+    "* **兼容性**：将 MindSpore 数据集转换为 Python `List`，使其天然支持索引访问（`__getitem__`），完美适配 Trainer 的数据分发逻辑。\n",
+    "* **训练加速**：避免了训练过程中频繁的磁盘 I/O 读取和实时预处理，数据直接从内存获取，极大提升训练吞吐量。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "10",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "print(\"Converting datasets to memory lists (Compatibility Bridge)...\")\n",
+    "\n",
+    "def ms_dataset_to_list(ms_dataset):\n",
+    "    data_list = []\n",
+    "    iterator = ms_dataset.create_dict_iterator(num_epochs=1, output_numpy=True)\n",
+    "    for item in iterator:\n",
+    "        data_list.append(item)\n",
+    "    return data_list\n",
+    "\n",
+    "# 执行转换\n",
+    "train_dataset_list = ms_dataset_to_list(train_dataset)\n",
+    "eval_dataset_list = ms_dataset_to_list(eval_dataset)\n",
+    "print(f\"Converted Train Size: {len(train_dataset_list)}\")\n",
+    "print(f\"Converted Eval Size: {len(eval_dataset_list)}\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "11",
+   "metadata": {},
+   "source": [
+    "## 4. 模型构建\n",
+    "\n",
+    "使用 `AutoModelForImageClassification` 加载预训练的 Swin-Tiny 模型。\n",
+    "- `num_labels`: 设置为 10（EuroSAT 的类别数）。\n",
+    "- `ignore_mismatched_sizes=True`: 允许加载时忽略分类头尺寸不匹配的问题（因为预训练模型是 1000 类 ImageNet，我们需要替换为 10 类），使用随机初始化的分类头。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "12",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from mindnlp.transformers import AutoModelForImageClassification\n",
+    "\n",
+    "model = AutoModelForImageClassification.from_pretrained(\n",
+    "    model_checkpoint,\n",
+    "    num_labels=len(labels),\n",
+    "    label2id=label2id,\n",
+    "    id2label=id2label,\n",
+    "    ignore_mismatched_sizes=True\n",
+    ")\n",
+    "\n",
+    "print(\"Model loaded successfully.\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "13",
+   "metadata": {},
+   "source": [
+    "## 5. 模型训练\n",
+    "\n",
+    "使用 MindSpore NLP 提供的 `Trainer` 接口进行训练。我们需要定义 `TrainingArguments` 来配置训练参数（学习率、Epoch 数、保存策略等）。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "14",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from mindnlp.transformers import Trainer, TrainingArguments\n",
+    "import evaluate\n",
+    "import numpy as np\n",
+    "\n",
+    "# 加载评估指标\n",
+    "metric = evaluate.load(\"accuracy\")\n",
+    "\n",
+    "# 定义评估函数\n",
+    "def compute_metrics(eval_pred):\n",
+    "    logits, labels = eval_pred\n",
+    "    predictions = np.argmax(logits, axis=-1)\n",
+    "    return metric.compute(predictions=predictions, references=labels)\n",
+    "\n",
+    "BATCH_SIZE = 32\n",
+    "# 配置训练参数\n",
+    "training_args = TrainingArguments(\n",
+    "    output_dir=\"swin_tiny_eurosat_finetune\", # 输出目录\n",
+    "    eval_strategy=\"epoch\",                   # 每个 epoch 评估一次\n",
+    "    save_strategy=\"epoch\",                   # 每个 epoch 保存一次\n",
+    "    learning_rate=5e-5,                      # 学习率\n",
+    "    per_device_train_batch_size=BATCH_SIZE,  # 训练 Batch Size\n",
+    "    per_device_eval_batch_size=BATCH_SIZE,   # 评估 Batch Size\n",
+    "    num_train_epochs=1,                      # 训练轮次\n",
+    "    weight_decay=0.01,                       # 权重衰减\n",
+    "    load_best_model_at_end=True,             # 训练结束后加载最佳模型\n",
+    "    metric_for_best_model=\"accuracy\",        # 以准确率作为最佳模型指标\n",
+    "    logging_steps=10,                        # 每 10 步打印日志\n",
+    "    save_total_limit=2                       # 最多保存 2 个 Checkpoint\n",
+    ")\n",
+    "\n",
+    "# 初始化 Trainer\n",
+    "trainer = Trainer(\n",
+    "    model=model,\n",
+    "    args=training_args,\n",
+    "    train_dataset=train_dataset_list,\n",
+    "    eval_dataset=eval_dataset_list,\n",
+    "    compute_metrics=compute_metrics,\n",
+    ")\n",
+    "\n",
+    "# 开始训练\n",
+    "trainer.train()"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "15",
+   "metadata": {},
+   "source": [
+    "## 6. 模型评估与保存\n",
+    "\n",
+    "训练完成后，我们在验证集上进行最终评估，并保存微调后的模型。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "16",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# 在验证集上评估\n",
+    "eval_metrics = trainer.evaluate()\n",
+    "print(f\"Evaluation Metrics: {eval_metrics}\")\n",
+    "\n",
+    "# 保存最终模型\n",
+    "trainer.save_model(\"output/swin-tiny-eurosat\")\n",
+    "print(\"Model saved to output/swin-tiny-eurosat\")"
+   ]
+  },
+  {
+   "cell_type": "markdown",
+   "id": "17",
+   "metadata": {},
+   "source": [
+    "## 7. 推理演示\n",
+    "\n",
+    "使用微调后的模型对一张新的网络图片进行分类预测。我们将展示如何使用 Pipeline API 快速进行推理。"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "id": "18",
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "from PIL import Image\n",
+    "import requests\n",
+    "from mindnlp.transformers import pipeline\n",
+    "\n",
+    "\n",
+    "# 下载一张测试图片（不在训练集中）\n",
+    "url = 'https://huggingface.co/nielsr/convnext-tiny-finetuned-eurostat/resolve/main/forest.png'\n",
+    "try:\n",
+    "    image = Image.open(requests.get(url, stream=True).raw)\n",
+    "except:\n",
+    "    # 如果网络无法访问，创建一个随机图片用于演示代码跑通\n",
+    "    print(\"Network error, using dummy image.\")\n",
+    "    image = Image.new('RGB', (224, 224), color='green')\n",
+    "\n",
+    "print(f\"Test Image Info: {image.format}, Size: {image.size}\")\n",
+    "\n",
+    "# 创建推理 Pipeline\n",
+    "# task: 图像分类\n",
+    "# model: 刚才训练好的模型对象\n",
+    "# image_processor: 预处理配置\n",
+    "pipe = pipeline(\"image-classification\", model=model, image_processor=image_processor)\n",
+    "\n",
+    "# 执行预测\n",
+    "result = pipe(image)\n",
+    "\n",
+    "# 打印结果\n",
+    "print(\"\\nPrediction Result:\")\n",
+    "for res in result:\n",
+    "    print(f\"Label: {res['label']}, Score: {res['score']:.4f}\")\n",
+    "\n",
+    "print(f\"Test Image Info: {image.format}, Size: {image.size}\")  "
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "Python (Ascend-Final)",
+   "language": "python",
+   "name": "my_analysis"
+  },
+  "language_info": {
+   "codemirror_mode": {
+    "name": "ipython",
+    "version": 3
+   },
+   "file_extension": ".py",
+   "mimetype": "text/x-python",
+   "name": "python",
+   "nbconvert_exporter": "python",
+   "pygments_lexer": "ipython3",
+   "version": "3.10.19"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 5
+}