## BinaryNet.tf
Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1. implementation in tensorflow
This is incomplete training example for BinaryNets using Binary-Backpropagation algorithm as explained in
"Binarized Neural Networks: Training Deep Neural Networks with Weights and Activations Constrained to +1 or -1,
on following datasets: Cifar10/100.
Note that in this folder I didn’t implemented (yet...) shift-base BN , shift-base AdaMax (instead I just use the vanilla BN and Adam).
Likewise, I use deterministic binarization and I don't apply the initialization coefficients from GLorot&Bengio 2010.
Finally "sparse_softmax_cross_entropy_with_logits" loss is used instead if the SquareHingeLoss.
## Data
This implementation supports cifar10/cifar100
## Dependencies
tensorflow version 1.2.1
## Training
* Train cifar10 model using gpu:
python main.py --model BNN_cifar10 --save BNN_cifar10 --dataset cifar10 --gpu True
* Train cifar10 model using cpu:
python main.py --model BNN_cifar10 --save BNN_cifar10 --dataset cifar10
## Results
Cifar10 should reach at least 88% top-1 accuracy
tensorflow-使用tensorflow实现的BNN算法.zip
需积分: 1 82 浏览量
2024-03-03
15:26:18
上传
评论
收藏 13KB ZIP 举报
Mopes__
- 粉丝: 1075
- 资源: 329
最新资源
- mysql查询语句汇总.zip
- typora-x64-v0.9.93,windows,64位,markdown工具,文本编辑器
- 基于C语言的neon_osd_Draw ARM Neon加速OSD点阵设计源码
- 目标检测-工地工人安全设备佩戴检测数据集-3500张图-+对应VOC-COCO-YOLO三种格式标签+数据集划分脚本+训练日志
- 运筹学运输问题综合实验的实验报告
- home - 副本.vue
- imageProcess.py
- typora-linux-v0.11.18-amd64,linux,64位,markdown工具,文本编辑器
- 基于Java的sky-take-out云霄外卖管理系统设计源码
- SAP-可用性检查应用测试.docx
资源上传下载、课程学习等过程中有任何疑问或建议,欢迎提出宝贵意见哦~我们会及时处理!
点击此处反馈