AI-BMT Platform — Python Submitter Interface

Last Updated: 2025-08-03

---

1. Environment

• ISA(Instruction Set Architecture) : ARM64(aarch64)
• OS : Ubuntu 22.04 LTS, 24.04 LTS
• Python Version: 3.8.X ~ 3.12.X supported

---

2. Build System Set-up

1. Install Packages

• Open a terminal and run the following commands to install CMake, g++ compiler, Ninja Build System, and EGL Library.

  sudo apt update
  sudo apt install cmake                     # CMake
  sudo apt install build-essential           # GCC, G++, Make
  sudo apt-get install ninja-build           # Ninja
  sudo apt install libgl1 libgl1-mesa-dev    # EGL and OpenGL
  sudo apt install unzip                     # unzip

2. Verify the Installation

• You can check the versions of the installed tools by running the following commands. If these commands return version information for each tool, the installation was successful.

  cmake --version
  gcc --version
  ninja --version
  dpkg -l | grep -E 'libgl1|libgl1-mesa-dev'

---

3. Project Description

This version of the AI-BMT Platform allows you to implement your submitter in Python by inheriting the provided abstract interface bmt.AI_BMT_Interface.
Once implemented, your model and preprocessing pipeline can be evaluated through the unified GUI interface, just like C++-based submitters.

You can directly modify the class SubmitterImplementation(bmt.AI_BMT_Interface) in main.py.
We also provide ONNX Runtime-based example scripts for Classification, Object Detection, and Semantic Segmentation in the example/ folder.

---

4. Submitter Development Guide

Required Interface

submitter must subclass bmt.AI_BMT_Interface and implement the following methods:

class SubmitterImplementation(bmt.AI_BMT_Interface):
    def Initialize(self, model_path: str) -> None:
        # Load and initialize your model here
        ...

    def convertToPreprocessedDataForInference(self, image_path: str) -> VariantType:
        # Perform image loading and preprocessing here
        ...

    def runInference(self, data: List[VariantType]) -> List[BMTResult]:
        # Perform inference and return results
        ...

Optional Interface

submitter can optionally provide hardware/system metadata using:

class SubmitterImplementation(bmt.AI_BMT_Interface):
    def getOptionalData(self) -> Optional_Data:
        data = Optional_Data()
        data.cpu_type = "Intel i7-9750HF"
        data.accelerator_type = "DeepX M1 (NPU)"
        data.submitter = "DeepX"
        data.cpu_core_count = "16"
        data.cpu_ram_capacity = "32GB"
        data.cooling = "Air"
        data.cooling_option = "Active"
        data.cpu_accelerator_interconnect_interface = "PCIe Gen5 x16"
        data.benchmark_model = "ResNet-50"
        data.operating_system = "Windows 10"
        return data

5. Start BMT

using following commands in AI_BMT_GUI_Submitter_Linux_ARM64_Python/ directory.

export LD_LIBRARY_PATH=$(pwd)/lib
python main.py