AI Computation Performance Improves and Rockchip Officially Launches Android 8.1 NNAPI SDK
2018-07-05Rockchip has officially launched the Android 8.1 Neural Networks API (NNAPI) Optimized SDK based on RK3399 Platform in order to provide the AI operational support featuring universal model and more powerful performance, which is suitable for various derivative and developed applications based on the mainstream model architecture, including face recognition, ADAS, goods identification and fatigue detection etc..
RK3399 features high performance, high scalability and versatility. CPU has applied the big.LITTLE architecture, Dual-core Cortex-A72 and Quad-core Cortex-A53, and its overall performance and power consumption have occupied a leading status technically. Meanwhile, GPU has applied the Quad-core ARM new-generation high-end image processor Mali-T860, and integrated much more bandwidth compression techniques to realize the excellent overall performance. In addition, Android 8.1 is able to completely invoke the neural neural network API (NNAPI) to realize the hardware acceleration, largely improving the AI Operational Performance of RK3399.
RK3399 SDK launched this time is a kind of GPU acceleration solution based on Android 8.1 NNAPI, which can provide corresponding acceleration support for related AI applications developed by means of Android NNAPI, and features the following four major advantages:
1. High Compatibility: Standard API can directly support various APK applications developed based on Android NNAPI;
2. Strong Universality: It can support numerous mainstream model architectures, and is suitable for various derivative and developed applications based on the mainstream model architecture, including face recognition, ADAS, goods identification and fatigue detection etc.;
3. Improved Performance: The real-time performance can be obtained among multiple tasks. For example, the highest frame rate can reach 23.2 when MobileNet is applied for image identification;
4. Lower Power Consumption: It can realize the high-efficient operation based on GPU, and the power consumption is only 1W when fully loaded
According to the APK Testing Data of image identification and target detection, the mainstream model’s performance is outstanding:
AI Computation has achieved an explosive growth, and Rockchip AI solution is widely applied to various industries including the image identification, the intelligent security, the intelligent drive, Automatic Speech Recognition, and the consumer electronics. However, Android 8.1 NNAPI Optimized SDK based on the RK3399 Platform is now officially launched to enhance the hard power of the AI solution Product Line.