{
  "name": "Swin-Transformer",
  "id": "ko3ugg54j400",
  "description": "Swin-Transformer是针对于图片处理设计的基于Transformer架构的神经网络。该网络针对原始Transformer迁移到图片端后计算量过大，复用困难的问题，提出了新的swin-block以代替原有的attention架构。模型以窗口的attention方式极大地减少了图像不同区域间的互相响应，同时也避免了部分冗余信息的产生。最终，模型在减少了大量计算量的同时，在不同的视觉传统任务上也有了效果的提升。",
  "category": "计算机视觉",
  "framework": [
    "PyTorch"
  ],
  "supportOs": [
    "OpenHarmony",
    "Linux"
  ],
  "computingPower": [
    "Hi3403V100 SVP_NNN",
    "Hi3403V100 NNN"
  ],
  "tags": [
    "分类"
  ],
  "repositoryUrl": "https://gitee.com/HiSpark/modelzoo/tree/master/samples/built-in/classification/Swin-Transformer",
  "licenseUrl": "https://github.com/microsoft/Swin-Transformer/blob/main/LICENSE",
  "downloads": [
    {
      "fileName": "swin.onnx",
      "variants": [
        "源模型 / 源模型元数据"
      ]
    },
    {
      "fileName": "swin_tiny_patch4_window7_224.pth",
      "variants": [
        "源模型 / 源模型元数据"
      ]
    },
    {
      "fileName": "swin_sim.onnx",
      "variants": [
        "源模型 / 源模型元数据"
      ]
    },
    {
      "fileName": "swin.om",
      "variants": [
        "编译模型 / OM 元数据 / A8W8"
      ]
    },
    {
      "fileName": "swin_dlite.om",
      "variants": [
        "编译模型 / OM 元数据 / FP16"
      ]
    },
    {
      "fileName": "SVP_NNN_PC_V1.0.6.0.tgz",
      "variants": [
        "附加资源 / 附加资源"
      ]
    },
    {
      "fileName": "swin_om-A8W8.om",
      "variants": [
        "编译模型 / 镜像补充 / A8W8"
      ]
    }
  ],
  "apiDetail": {
    "createdBy": 139036723332644,
    "creationDate": "2026-08-26 16:38:28",
    "creationUserCN": "y-0089",
    "lastUpdatedBy": null,
    "lastUpdateDate": "2026-09-01 16:13:06",
    "lastUpdateUserCN": "y-0089",
    "rowIdx": -1,
    "id": "ko3ugg54j400",
    "name": "Swin-Transformer",
    "isBeta": 0,
    "betaVersionDesc": "",
    "description": "Swin-Transformer是针对于图片处理设计的基于Transformer架构的神经网络。该网络针对原始Transformer迁移到图片端后计算量过大，复用困难的问题，提出了新的swin-block以代替原有的attention架构。模型以窗口的attention方式极大地减少了图像不同区域间的互相响应，同时也避免了部分冗余信息的产生。最终，模型在减少了大量计算量的同时，在不同的视觉传统任务上也有了效果的提升。",
    "parentId": "h6s4m1ogto00",
    "coverImageId": 1701485580845059,
    "coverImageUrl": "https://openxinhuo-board-image.obs.cn-east-3.myhuaweicloud.com/1701485580845059%2Fswint.png",
    "modelEffectId": null,
    "modelEffectUrl": "",
    "computerVersion": [
      "分类"
    ],
    "naturalLanguageProcess": [],
    "multimodal": [],
    "video": [],
    "framework": [
      "PyTorch"
    ],
    "modelRepository": "https://gitee.com/HiSpark/modelzoo/tree/master/samples/built-in/classification/Swin-Transformer",
    "originModel": [
      {
        "id": "1706935124623362",
        "name": "swin.onnx",
        "url": null,
        "size": "115360652"
      },
      {
        "id": "1706935036477443",
        "name": "swin_tiny_patch4_window7_224.pth",
        "url": null,
        "size": "114342173"
      },
      {
        "id": "1706935070031874",
        "name": "swin_sim.onnx",
        "url": null,
        "size": "115268484"
      }
    ],
    "originModelLink": "",
    "dataSet": "https://image-net.org/",
    "modelLicense": "https://github.com/microsoft/Swin-Transformer/blob/main/LICENSE",
    "detailParams": [
      {
        "name": "输入",
        "value": "224x224"
      },
      {
        "name": "参数量",
        "value": "28.798M"
      },
      {
        "name": "计算量",
        "value": "9.567GFLOPs"
      }
    ],
    "quickStart": {
      "url": "https://gitee.com/HiSpark/modelzoo/tree/master/samples/built-in/classification/Swin-Transformer",
      "markDownUrl": "",
      "developLanguage": [
        {
          "language": "C++",
          "context": "{\"ops\":[{\"attributes\":{\"size\":\"16px\",\"color\":\"#333333\",\"line-height\":\"inherit\"},\"insert\":\"Swin-Transformer可以通过以下代码完成快速推理, 该代码仅展示主要流程，完整实现参考src/main.cpp文件和readme\"},{\"attributes\":{\"text-indent\":\"0px\"},\"insert\":\"\\n\"},{\"insert\":\"#include \\\"model.h\\\"\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"#include \\\"log.h\\\"\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\\n\"},{\"insert\":\"using namespace Infer;\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\\n\"},{\"insert\":\"int main()\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"{\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    EnvInit();\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    std::string omModelPath = \\\"/path/to/model.om\\\"; // Swin-Transformer 模型文件路径 \"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    std::string imagePath = \\\"/path/to/image.jpg\\\"; // 输入图片路径\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    std::unique_ptr<Model> model = std::make_unique<Model>();\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    if (model->Load(omModelPath, ModelType::SwinT) != 0) {\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"        LOG(ERROR) << \\\"fail to load model\\\";\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"        return -1;\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    }\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    auto ret = model->Infer(imagePath, FileType::SingelImageFile);\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    if (ret.size() == 0) {\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"        LOG(ERROR) << \\\"fail to infer model\\\";\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"        model->Unload();\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"        return -1;\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    }\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    if (model->Unload() != 0) {\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"        LOG(ERROR) << \\\"fail to unload model\\\";\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"        return -1;\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    }\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    EnvDeinit();\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"    return 0;\"},{\"attributes\":{\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"insert\":\"}\"},{\"attributes\":{\"text-indent\":\"0px\",\"code-block\":\"plain\"},\"insert\":\"\\n\"},{\"attributes\":{\"size\":\"16px\",\"color\":\"#333333\",\"line-height\":\"inherit\"},\"insert\":\"备注：头文件和动态库位于\"},{\"attributes\":{\"size\":\"16px\",\"color\":\"#4183c4\",\"line-height\":\"inherit\",\"link\":\"https://gitee.com/HiSpark/modelzoo/tree/master/samples/common\"},\"insert\":\"/samples/common\"},{\"attributes\":{\"size\":\"16px\",\"color\":\"#333333\",\"line-height\":\"inherit\"},\"insert\":\"目录下，编译配置参考\"},{\"attributes\":{\"size\":\"16px\",\"color\":\"#4183c4\",\"line-height\":\"inherit\",\"link\":\"https://gitee.com/HiSpark/modelzoo/blob/master/samples/built-in/classification/Swin-Transformer/src/CMakeLists.txt\"},\"insert\":\"文件\"},{\"attributes\":{\"text-indent\":\"0px\"},\"insert\":\"\\n\"}]}"
        }
      ]
    },
    "status": "released",
    "currentHandler": "",
    "currentHandlerName": "",
    "jsonPath": "https://gitee.com/HiSpark/modelzoo-dev/blob/master/samples/built-in/classification/Swin-Transformer/Swin-Transformer.json",
    "modelAdaptor": [
      {
        "createdBy": null,
        "creationDate": null,
        "creationUserCN": null,
        "lastUpdatedBy": null,
        "lastUpdateDate": null,
        "lastUpdateUserCN": null,
        "rowIdx": -1,
        "id": "i8ttm5k1tc00",
        "name": "Hi3403V100 SVP_NNN",
        "modelId": "ko3ugg54j400",
        "modelName": "Swin-Transformer",
        "supportNames": [
          "A8W8"
        ],
        "toolkit": [
          {
            "name": "CANN工具",
            "url": "https://hispark-obs.obs.cn-east-3.myhuaweicloud.com/SVP_NNN_PC_V1.0.6.0.tgz",
            "desc": "Al异构计算架构；提升计算效率的关键平台",
            "imgId": "cann"
          },
          {
            "name": "编译工具链",
            "url": "https://gitee.com/HiSpark/pegasus/blob/Beta-v0.9.1/docs/Hi3403V100%E7%8E%AF%E5%A2%83%E6%90%AD%E5%BB%BA%E6%8C%87%E5%8D%97/Hi3403V100%E7%8E%AF%E5%A2%83%E6%90%AD%E5%BB%BA%E6%8C%87%E5%8D%97.md#241%E5%AE%89%E8%A3%85clang%E4%BA%A4%E5%8F%89%E7%BC%96%E8%AF%91%E5%99%A8",
            "desc": "高效编译，精准适配；AI性能优化，应用流畅运行",
            "imgId": "tool"
          },
          {
            "name": "SDK",
            "url": "https://gitee.com/HiSpark/ss928v100_clang/tree/Beta-v0.9.1/",
            "desc": "稳定、易用的设计；支撑客户快速产品量产",
            "imgId": "sdk"
          }
        ],
        "supportOs": [
          "OpenHarmony",
          "Linux"
        ],
        "supportQuantify": [
          {
            "createdBy": 139036723332644,
            "creationDate": "2026-08-26 16:38:28",
            "creationUserCN": "y-0089",
            "lastUpdatedBy": 139036723332644,
            "lastUpdateDate": "2026-08-26 16:38:28",
            "lastUpdateUserCN": "y-0089",
            "rowIdx": -1,
            "id": "ko3uggecj400",
            "name": "A8W8",
            "computingId": "i8ttm5k1tc00",
            "computingName": "Hi3403V100 SVP_NNN",
            "omOfflineModelUrl": null,
            "omOfflineModelId": 1712880949002241,
            "omOfflineModelSize": "30888496",
            "omOfflineModelName": "swin.om",
            "omOfflineModel": [
              {
                "id": "1734292545798147",
                "name": "swin.om",
                "url": null,
                "size": "30870482"
              }
            ],
            "omOfflineModelLink": null,
            "releaseTime": "2025-11-18",
            "boardOs": "OpenHarmony",
            "modelLicense": "https://gitee.com/HiSpark/modelzoo/tree/master/samples/built-in/classification/Swin-Transformer/LICENSE",
            "modelPerformance": [
              {
                "performanceValue": "25.675",
                "unit": "耗时（ms）",
                "desc": ""
              },
              {
                "performanceValue": "38.949",
                "unit": "性能（fps）",
                "desc": ""
              },
              {
                "performanceValue": "295.711",
                "unit": "单帧内存带宽（MB）",
                "desc": ""
              },
              {
                "performanceValue": "38.176",
                "unit": "内存（MB）",
                "desc": ""
              }
            ],
            "deleted": 0
          }
        ],
        "deleted": 0
      },
      {
        "createdBy": null,
        "creationDate": null,
        "creationUserCN": null,
        "lastUpdatedBy": null,
        "lastUpdateDate": null,
        "lastUpdateUserCN": null,
        "rowIdx": -1,
        "id": "i8ttm5k1tc01",
        "name": "Hi3403V100 NNN",
        "modelId": "ko3ugg54j400",
        "modelName": "Swin-Transformer",
        "supportNames": [
          "A8W8",
          "FP16"
        ],
        "toolkit": [
          {
            "name": "CANN工具",
            "url": "",
            "desc": "5.30.t11.7.b110； (请联系FAE获取)",
            "imgId": "cann"
          },
          {
            "name": "编译工具链",
            "url": "",
            "desc": "aarch64-mix210-linux-gcc；(请联系FAE获取)",
            "imgId": "tool"
          },
          {
            "name": "SDK",
            "url": "",
            "desc": "SS928 V100R001C02SPC022； (请联系FAE获取)",
            "imgId": "sdk"
          }
        ],
        "supportOs": [
          "Linux"
        ],
        "supportQuantify": [
          {
            "createdBy": 139036723332644,
            "creationDate": "2026-08-26 16:38:28",
            "creationUserCN": "y-0089",
            "lastUpdatedBy": 139036723332644,
            "lastUpdateDate": "2026-08-26 16:38:28",
            "lastUpdateUserCN": "y-0089",
            "rowIdx": -1,
            "id": "ko3uggfsj400",
            "name": "FP16",
            "computingId": "i8ttm5k1tc01",
            "computingName": "Hi3403V100 NNN",
            "omOfflineModelUrl": null,
            "omOfflineModelId": 1712880995074050,
            "omOfflineModelSize": "62380441",
            "omOfflineModelName": "swin_dlite.om",
            "omOfflineModel": [
              {
                "id": "1712880995074050",
                "name": "swin_dlite.om",
                "url": null,
                "size": "62380441"
              }
            ],
            "omOfflineModelLink": null,
            "releaseTime": "2025-11-18",
            "boardOs": "Linux",
            "modelLicense": "https://gitee.com/HiSpark/modelzoo/tree/master/samples/built-in/classification/Swin-Transformer/LICENSE",
            "modelPerformance": [
              {
                "performanceValue": "101.97",
                "unit": "耗时（ms）",
                "desc": "性能工具。"
              },
              {
                "performanceValue": "9.81",
                "unit": "性能（fps）",
                "desc": "性能工具。"
              },
              {
                "performanceValue": "494.766",
                "unit": "单帧内存带宽（MB）",
                "desc": ""
              },
              {
                "performanceValue": "215.668",
                "unit": "内存（MB）",
                "desc": ""
              }
            ],
            "deleted": 0
          }
        ],
        "deleted": 0
      }
    ],
    "saveType": null,
    "deleteType": null,
    "latest": "Y",
    "deleted": 0,
    "modelPhase": "released",
    "remark": null,
    "fileInfo": null,
    "reviewType": null,
    "owner": "y-0089",
    "ownerBy": 139036723332644,
    "optional": null,
    "optionalList": [
      "0e900d99dee8461b8"
    ],
    "optionalBy": null,
    "downloadNum": 53,
    "collectNum": null,
    "isCollect": null
  }
}