---
title: 开源网络模型分解数据参考
description: "精度指标：分类：top1 ACC(%)，检测: mAP(%)，分割: DSC(%)。fine-tune学习率从原始学习率的0.1倍开始降低。"
url: https://www.hiascend.com/document/detail/zh/canncommercial/latest/devaids/amct/atlasamct_16_0249.html
sourcePath: /source/zh/canncommercial/900/devaids/amct/atlasamct_16_0249.html
indexId: 2349823165e144537d50def664fa82f2c958e6589675a4156694cba1136ba1ff65
---
# 开源网络模型分解数据参考

精度指标：分类：top1 ACC(%)，检测: mAP(%)，分割: DSC(%)。fine-tune学习率从原始学习率的0.1倍开始降低。

| 模型 | 任务类型 | 数据集 | 基线精度 | 分解后精度 | 分解后fine\-tune精度 |
| --- | --- | --- | --- | --- | --- |
| ResNet18 | 分类 | ImageNet | 70.66 | 44.02 | 70.34 |
| ResNet34 | 分类 | ImageNet | 74.2 | 54.92 | 74.15 |
| ResNet50 | 分类 | ImageNet | 75.6 | 73.64 | 75.91 |
| ResNet101 | 分类 | ImageNet | 78.52 | 76.97 | 78.24 |
| InceptionV3 | 分类 | ImageNet | 77.98 | 76.95 | 77.78 |
| SSD | 检测 | coco2017 | 27.2 | 24.2 | 27.9 |
| faster\-rcnn | 检测 | coco2017 | 32.5 | 31 | 32.2 |
| mask\-rcnn | 检测 | coco2017 | 37.9 | 36.8 | 38 |
| UNet | 分割 | SSTEM | 87.63 | 85.05 | 87.57 |