Skip to content

1st Place Solution for Efficient and Accurate Models towards Practical Deep Learning

License

Notifications You must be signed in to change notification settings

LingyvKong/UP-AAAI2023challenge-1st

Repository files navigation

AAAI2023 实用AI挑战赛冠军方案分享,比赛链接

1st Place Solution for Efficient and Accurate Models towards Practical Deep Learning

1. 数据集准备:

把官方data.zip中的图片解压到dataset/data/中;

或者把dataset/train_plus和dataset/val复制到你的data路径,并修改yaml中的数据集路径

这是我们基于官方图片生成的少量补充图片,详细描述见技术报告,你也可以通过以下方式自己生成:

  • 修改脚本dataview.py中的路径并运行main_split()函数,划分训练集和验证集
  • 运行脚本dataview.py中的main_train_plus()函数,生成train_plus补充训练集

2. 训练:

将 resnet_qavgg.py 第二十行的deploy设置为False

运行如下指令进行训练:

python train.py --config configs/local_AAAI/qavgg.yaml

3. 部署:

将resnet_qavgg.py 第二十行的deploy = False改为True

运行如下指令转换onnx:

python toonnx.py --config configs/local_AAAI/qavgg.yaml --save_prefix 128_5conv --input_size 3x28x28

4. 其他代码说明:

The code is highly borrowed from: https://github.com/ModelTC/United-Perception

United Perception 文档

up 官方文档

About

1st Place Solution for Efficient and Accurate Models towards Practical Deep Learning

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published