提交 8270b151 编写于 作者: Z Zero King

Update README.md

上级 70966273
<div align="center"> <div align="center">
<img src="docs/mace-logo.png" width = "400" alt="MACE" /> <img src="docs/mace-logo.png" width="400" alt="MACE" />
</div> </div>
...@@ -32,16 +32,16 @@ targets: ...@@ -32,16 +32,16 @@ targets:
introduced to allow better preemption for the UI rendering task. introduced to allow better preemption for the UI rendering task.
* Memory usage and library footprint * Memory usage and library footprint
* Graph level memory allocation optimization and buffer reuse are supported. * Graph level memory allocation optimization and buffer reuse are supported.
The core library tries to keep minium external dependencies to keep the The core library tries to keep minimum external dependencies to keep the
library footprint small. library footprint small.
* Model protection * Model protection
* Model protection is the highest priority feature from the beginning of * Model protection is the highest priority feature from the beginning of
the design. Various techniques are introduced like coverting models to C++ the design. Various techniques are introduced like converting models to C++
code and literal obfuscations. code and literal obfuscations.
* Platform coverage * Platform coverage
* A good coverage of recent Qualcomm, MediaTek, Pinecone and other ARM based * A good coverage of recent Qualcomm, MediaTek, Pinecone and other ARM based
chips. CPU runtime is also compatible with most POSIX systems and chips. CPU runtime is also compatible with most POSIX systems and
archetectures with limited performance. architectures with limited performance.
## Getting Started ## Getting Started
* [Introduction](https://mace.readthedocs.io/en/latest/getting_started/introduction.html) * [Introduction](https://mace.readthedocs.io/en/latest/getting_started/introduction.html)
...@@ -51,12 +51,12 @@ targets: ...@@ -51,12 +51,12 @@ targets:
## Performance ## Performance
[MACE Model Zoo](https://github.com/XiaoMi/mace-models) contains [MACE Model Zoo](https://github.com/XiaoMi/mace-models) contains
several common neural networks and models which will be built daily against a list of mobile several common neural networks and models which will be built daily against a list of mobile
phones. The benchmark result can be found in the CI result page. phones. The benchmark results can be found in the CI result page.
## Communication ## Communication
* GitHub issues: bug reports, usage issues, feature requests * GitHub issues: bug reports, usage issues, feature requests
* Mailing list: [mace-users@googlegroups.com](mailto:mace-users@googlegroups.com) * Mailing list: [mace-users@googlegroups.com](mailto:mace-users@googlegroups.com)
* Google groups: https://groups.google.com/forum/#!forum/mace-users * Google Groups: https://groups.google.com/forum/#!forum/mace-users
* QQ群: 756046893 * QQ群: 756046893
## Contributing ## Contributing
...@@ -69,7 +69,7 @@ please refer to [the contribution guide](https://mace.readthedocs.io/en/latest/d ...@@ -69,7 +69,7 @@ please refer to [the contribution guide](https://mace.readthedocs.io/en/latest/d
[Apache License 2.0](LICENSE). [Apache License 2.0](LICENSE).
## Acknowledgement ## Acknowledgement
MACE depends on several open source projects located in MACE depends on several open source projects located in the
[third_party](third_party) directory. Particularly, we learned a lot from [third_party](third_party) directory. Particularly, we learned a lot from
the following projects during the development: the following projects during the development:
* [Qualcomm Hexagon NN Offload Framework](https://source.codeaurora.org/quic/hexagon_nn/nnlib): the Hexagon DSP runtime * [Qualcomm Hexagon NN Offload Framework](https://source.codeaurora.org/quic/hexagon_nn/nnlib): the Hexagon DSP runtime
...@@ -82,4 +82,4 @@ the following projects during the development: ...@@ -82,4 +82,4 @@ the following projects during the development:
practices from these projects. practices from these projects.
Finally, we also thank the Qualcomm, Pinecone and MediaTek engineering teams for Finally, we also thank the Qualcomm, Pinecone and MediaTek engineering teams for
their helps. their help.
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册