作者:手机用户2602884633 | 来源:互联网 | 2023-08-06 13:32
-
- 简介
- 文档与性能
- 网络与模型能力
- 生态与维护
- 框架搭建与应用
- 架构
- 1 TensorFlow
- 2 Caffe
- 3 MXNet
- 4 Paddle
- 5 CNTK
- 总结
1.简介
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcxNTQzODQxP3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
2.文档与性能
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcxNjIxODQ2P3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
3.网络与模型能力
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcxNzA3NDUyP3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
4.生态与维护
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcxNzMzNzY0P3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
5.框架搭建与应用
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcxNzQ3NDk5P3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
6.架构
6.1 TensorFlow
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcyMzM5NDAyP3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
6.2 Caffe
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcyMzU2NTEyP3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
6.3 MXNet
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcyNDI4ODI5P3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
6.4 Paddle
主要从以下几个方面入手:多机并行架构、多GPU并行架构、sequence序列模型以及大规模稀疏训练;
6.5 CNTK
![这里写图片描述](https://www.#.com/go/aHR0cDovL2ltZy5ibG9nLmNzZG4ubmV0LzIwMTcwNTAzMTcyNDUyNTgwP3dhdGVybWFyay8yL3RleHQvYUhSMGNEb3ZMMkpzYjJjdVkzTmtiaTV1WlhRdlpYUmxjbTVwZEhreE1URTRYdz09L2ZvbnQvNWE2TDVMMlQvZm9udHNpemUvNDAwL2ZpbGwvSTBKQlFrRkNNQT09L2Rpc3NvbHZlLzcwL2dyYXZpdHkvU291dGhFYXN0)
7.总结
1). 有关图像的问题使用caffe很方便,训练只需要写prototxt;
2). Caffe是目前产品化最多的库;
3). 应该多关注TensorFlow,毕竟有Google这样的亲爹;
4). 用TensorBoard去观察训练的状态;
5). 可以了解和学习一下mxnet,对显存利用率高;