Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

Appearance settings

powerli2002/ChineseFoodNet-Resnet50

Open more actions menu

Repository files navigation

Chinese traditional food picture classification

主要功能为使用Resnet50在ChineseFoodNet上进行训练,搭建分类网络。

About ChineseFoodNet

这个是一个关于中国传统食物图片的数据集。其中包含了208类食物,数据集图片总量大小约20G,数量约18W,关于数据集的详细描述可以参见ChineseFoodNet

2024.12.24 Update

  1. 添加了原论文中的数据增强功能。直接输入图片训练无法达到数据集精度,依据论文中的数据增强方法进行训练,(输入前进行随机裁剪和50%水平翻转),有效抑制了模型过拟合趋势。
  2. 添加了融合cbam 进行模型训练的模块,包括训练代码train_cbam.py,网络实现代码resnet_cbam.py。本次选择在BottleNeck的三层卷积层后依次加入通道注意力和空间注意力模块。(添加cbam模块后可以正常加载预训练权重)
  3. 在最新的训练代码中,加入了对wandb的适配。

Performance Comparison

Model Accuracy Val Top 1 Val Top 5 Test Top 1 Test Top 5
ResNet50 68.14% 86.51% 68.39% 86.61%
ResNet50 + DA 78.88% 93.06% 79.23% 93.49%
ResNet50 + DA + CBAM 79.67% 93.71% 80.25% 93.78%

Specific implementation

  1. 实现了使用Resnet50训练,并达到测试集与验证集上top1 70%,top5 90%的精度。
  2. 模型具有保存和读取功能。
  3. 添加了计算top1和top5 accuracy的函数,可计算train,test,valid等数据集上的准确率

Usage

PyTorch实现版本为1.8.1。

1.将数据集下载到ChineseFoodNet文件夹:

数据集下载: https://pan.baidu.com/s/19lPkSGhMwe5QLLXHNOu-Zw?pwd=7rur

image-20240909100841295

2.项目根目录创建model_data文件夹,权重放入model_data文件夹下。

训练权重下载: 链接: https://pan.baidu.com/s/1szYn5OTT01eNCkwr2fWTQA?pwd=w2uj

此权重实现效果:

Accuracy Top 1 Top 5
Train 97.28% 99.83%
Valid 68.76% 90.95%
Test 68.93% 91.09%

3.cal_top5acc.py ,cal_top1acc.py 可切换验证集,测试集。修改dataset_test即可。

References

  • Chen X, Zhu Y, Zhou H, et al. Chinesefoodnet: A large-scale image dataset for chinese food recognition[J]. arXiv preprint arXiv:1705.02743, 2017.
  • He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[C]//Proceedings of the IEEE conference on computer vision and pattern recognition. 2016: 770-778.
  • https://github.com/paradiseDust/ChineseFoodNet-EffiNet-L2
  • Woo S, Park J, Lee J, et al. CBAM: Convolutional Block Attention Module[J]. arXiv preprint arXiv:1807.06521, 2018.

About

Training code for ChineseFoodNet dataset. 对ChineseFoodNet数据集的训练代码

Topics

Resources

Stars

Watchers

Forks

Releases

Packages

Used by

Contributors

Languages

Morty Proxy This is a proxified and sanitized view of the page, visit original site.