Are you sure you want to delete this task? Once this task is deleted, it cannot be recovered.
You can not select more than 25 topics
Topics must start with a chinese character,a letter or number, can include dashes ('-') and can be up to 35 characters long.
stillwaters
93b71e6101
|
2 years ago | |
|---|---|---|
| precision_tool | 2 years ago | |
| LICENSE | 2 years ago | |
| RCA_net.py | 2 years ago | |
| README.md | 2 years ago | |
| data_loader.py | 2 years ago | |
| dl_datasets.sh | 2 years ago | |
| main.py | 2 years ago | |
| mode.py | 2 years ago | |
| ops.py | 2 years ago | |
| train_testcase.sh | 2 years ago | |
| util.py | 2 years ago | |
README.md
RCAN_TensorFlow_for_NPU
项目说明
本项目是RCAN的TensorFlow版本在昇腾NPU实现的代码。
参考论文
Image Super-Resolution Using Very Deep Residual Channel Attention Networks
参考实现
https://github.com/dongheehand/RCAN-tf
Requirement
tensorflow==1.15.0
cann==5.0.2
numpy==1.17.5
scikit-image==0.17.2
默认配置
- 网络结构
RCAN×2 - 训练超参（单卡）
batch_size:32
learning_rate(LR):2*1e-4
max_steps:1e6
安装教程
使用git clone {repository url}
或者下载本项目到本地并解压缩
数据集
使用说明
- 数据集准备
在项目的工作目录下，在命令行执行如下脚本一步完成数据集下载和更改名称bash dl_datasets.sh - 训练和推理
在完成数据集准备工作之后，在命令行执行如下命令进行训练和训练后的推理，得到精度最好的模型bash train_testcase.sh
第四届中国软件开源创新大赛·赛道二：任务挑战赛（模型王者挑战赛）：基于华为Ascend 910，利用Tensorflow 1.15.0 实现RCAN图像超分网络，数据集：训练集：DIV2K 测试集：Set14，最终精度：PSNR=34.2681，SSIM：0.9236 性能：0.32s/step
Python Shell HAProxy