Competition Rules Data and Evaluation Leaderboard FAQs
Using SOPHGO TPU Submission Requirements Evaluation

Using SOPHGO TPU

  • Hardware Platform

Shaolin Pi Development Board

Shaolin Pi development board is a high-performance platform based on BM1684, providing around 20 TOPS of computing power. With BM1684 as its core component, the board features a fully autonomous and controllable processor, delivering exceptional computing power and multi-channel video encoding and decoding capabilities. It supports 3 mini-PCIe and 4 USB interfaces, allowing for the expansion of various peripheral modules. It optimizes configurations according to scene requirements, balancing cost-effectiveness, energy efficiency, and functionality. The hardware ecosystem is extensive, accommodating diverse peripheral connections. It boasts support for a rich software development ecosystem, compatible with mainstream deep learning frameworks. The 'Shaolin Pi' core board can expand connectivity to screens, keyboards, mice, cameras, headphones, VR, and other devices. It enables the creation of an all-encompassing edge computing workstation for various experiments. Additionally, it can be embedded into unmanned vehicles and drones, enabling edge computing for mobile terminals. Leveraging the Shaolin Pi development board, our company has developed the KT001 smart car and S550 deep learning drone. Participants can also request the deep learning car as a hardware platform to complete their own parameterized works for the competition.

 

  • Software Platform

SophonSDK is a deep learning SDK tailored by SOPHGO for its independently developed processor. It encompasses capabilities such as model optimization and efficient runtime support required during the neural network inference phase, providing an easy-to-use and efficient end-to-end solution for deep learning application development and deployment.

 

SophonSDK consists of a Compiler and Library: The Compiler is responsible for offline compilation and optimization of neural network models trained under third-party deep learning frameworks, generating the required BModel for the final runtime. It currently supports Caffe, Darknet, MXNet, ONNX, PyTorch, PaddlePaddle, TensorFlow, among others. The Library includes BM-OpenCV, BM-FFmpeg, BMCV, TPURuntime, BMLib, and other libraries, which drive hardware components like VPP, VPU, JPU, TPU, enabling video/image encoding/decoding, image processing, tensor operations, model inference, and other functions for users in deep learning application development.

 

  • SOPHGO provides SDK-related resources for participants to learn:

  1. Documentation Center: https://developer.sophgo.com/site/index/material/30/all.html

  2. Video Tutorials: https://developer.sophgo.com/site/index/course/all/all.html

  3. Development Guide: https://sophgo-doc.gitbook.io/sophonsdk3

Submission Requirements

  • Presentation of the System Solution PPT
  • Introduction of the Solution and Functional Demonstration Video
  • Document on Solution Design and Algorithm Implementation
  • Engineering source code with clear explanatory comments

Evaluation

The total score for the project is 100 points, with specific scores allocated as follows:

  • Design Phase (50 points), comprising:
  1. Innovation in Edge Computing Application Scenarios (10 points)
  2. Utilization of Shaolin Pi Development Board in Design Process (10 points)
  3. System Functionality and Scalability (10 points)
  4. Software Algorithm Performance and Innovation (10 points)
  5. Utilization of Compiler (10 points)

 

  • Implementation Phase (30 points), comprising:
  1. Construction of Hardware Platform (10 points)
  2. Feature Implementation and Refinement (20 points)

 

  • Project and Output Formats (20 points), comprising:
  1. Hardware Circuits, Detailed Design Documents, and Software Code (15 points)
  2. System Demonstration (5 points)