Xiaomi MACE, an Open Source AI Project, announced yesterday at GitHub. The company now inviting developers to participate. This platform was actually released on December 15, 2017, by Xioami but now it’s opened for the testing process. It is designed for mobile chip optimization through AI and supports heterogeneous computing platform (CPU/GPU/DSP), TensorFlow/Caffe model, and multiple SoCs including Qualcomm, MediaTek, and other chips. Now it’s in the development process and will work for some internal firmware features like AI Portrait mode, scene recognition etc.
Xioami MACE is based on a deep machine learning inference framework that is highly optimized for the mobile heterogeneous computing platform. You can check out the demo here.
You May Also Like:
The Xiaomi MACE design is focused on below-mentioned things:
The runtime is highly optimized with NEON, OpenCL, Hexagon, and Winograd algorithm is introduced to speed up the convolution operations. Besides the fast inference speed, the initialization part is also intensively optimized to be faster.
Chip dependent power options like big.LITTLE scheduling, Adreno GPU hints are included as advanced APIs.
UI responsiveness guarantee is sometimes obligatory when running a model. The mechanism like automatically breaking OpenCL kernel into small units is introduced to allow better preemption for the UI rendering task.
Memory usage and library footprint
Graph level memory allocation optimization and buffer reuse are supported. The core library tries to keep minimum external dependencies to keep the library footprint small.
Model protection is the highest priority feature from the beginning of the design. Various techniques are introduced like converting models to C++ code and literal obfuscations.
A good coverage of recent Qualcomm, MediaTek, Pinecone and other ARM-based chips. CPU runtime is also compatible with most POSIX systems and architectures with limited performance.
Xioami MACE Performance
Xioami MACE Model Zoo contains several common neural networks and processes which will be built for daily mobile tasks. The benchmark results can be found through the CI result. Xioami MACE depends on several open source projects like Qualcomm Hexagon NN Offload Framework, TensorFlow/Caffe model, SNPE, ARM ComputeLibrary, ncnn etc.
GitHub gives a special thanks to the Qualcomm, Pinecone and MediaTek engineering teams for their extended support in the Xiaomi MACE development. However, GitHub welcoming all the contributors to join this platform and test this to find out bugs, feature requests, or any other issue.
According to the Vice President of Xioami, Cui Baoqiu, Artifical Intelligence, and the Cloud Platform will enhance the open source project as an important part of Xioami’s Engineering platform. Thanks to the Qualcomm, Pinecone, MTK, GitHub, and TensorFlow to make a better development. The Cloud Computing Machine Learning Platform will gather and store a huge amount of data without any issue. Xiaomi hoping that this Machine Learning Mobile AI Compute Engine (Xiaomi MACE) will empower the Chinese Artificial Intelligence Industry towards the Science and Technology field.