AI Artifact Package

• Definition
  • Catalog of AI Artifacts
• Meta Data
  • Metadata for AI Artifacts
• Create Package

AI Challenge

• Definition
  • Schema for Bonseyes Challenges
• Create an AI Challenge
  • Creating the challenge metadata
  • Defining the challenge interface
    • Bonseyes Interface Description
    • Default Interfaces
      • Image classification
      • Object detection
      • Face recognition
      • Image segmentation
      • Audio classification
      • Signal classification
  • Defining the challenge training and sample data (Deprecated)
  • Defining the evaluation procedure (Deprecated)
• Upload Challenge

AI Asset

• Definition
• Create AI Asset
  • Setup
    • Gitlab setup
    • Local setup
    • HPC setup
    • GitLab runner setup on HPC instance
    • Start new project
    • Local development workflow
  • I. Data
  • II. Train
    • Get data
    • Config file
    • train script
  • III. Model Catalog
    • Models nomenclature and storage
    • Model summary
  • IV. Algorithm
    • Algorithm base classes
    • Algorithm inherited classes
    • LPDNN Algorithm
    • Challenge Interface
  • V. Export
    • ONNX export
      • torch2onnx export
      • tf2onnx export
    • TensorRT export
    • torch2trt export
    • onnx2lpdnn export
    • All export
  • VI. Optimize
    • Post Training Quantization (PTQ)
      • TensorRT PTQ
      • ONNXRuntime PTQ
      • torch2trt PTQ
      • LPDNN PTQ
      • All PTQ
    • Quantization Aware Training (QAT)
      • QAT Tools
      • QAT process
  • VII. Process
    • Image Process
    • Video Process
    • Camera Process
    • HTTP Worker Process
    • LPDNN process
      • Image process
      • Video Process
      • Camera Process
  • VIII. Benchmark
    • Models benchmark
    • LPDNN benchmark
    • Generate graphs
  • IX. Utils
    • environment_info
    • gstreamer_pipelines
    • meter
    • hardware_info
  • X. Testing
  • XI. AI Asset CLI Integration
  • XII. Documentation

Data Tool

• Definition
  • Scope
  • Data Traceability with Versioning and Tagging
  • Datatool Output
  • Post Data-tooling Benefits
• Pre-requisites
  • Access to the Source Dataset
  • Complete Documentation of the Source Dataset
  • JSON data model from dataset documentation
  • Technical Requirements for the Development Environment
• Create Datatool
  • Datatool setup (How to use Template):
    • Step 1:
    • Step 2:
    • Step 3:
    • Step 4:
  • Datatool Structure
  • Data Model
  • Datatool Implementation
    • Step 1: Update Config Files
    • Step 2: Update Config Schema Files
    • Step 3: Add Python Dependency Modules
    • Step 4: Implement Python Data Model for the Annotations
    • Step 5: Create Unit Tests for the Data model
    • Step 6: Complete Source Dataset Parser Implementation
    • Step 7: Complete Datatool Patch Implementation (Optional)
    • Step 8: Complete Visualize Annotations Implementation
    • Step 9: Update User Documentation
    • Step 10: Generate and Add EDA Report
    • Step 11: Build, Test and Push the Docker Image
    • Step 12: Complete Example Dataloader Implementation

Evaluation Tool

• Definition
  • Bonseyes AI App Evaluation Protocol Metadata
• Create Evaluation Tool
  • Create the evaluation tool script
  • Dockerize the evaluation tool script
  • Create the evaluation tool metadata

Platform

• BSP Structure
  • General
  • Platform sources
  • Platform support
  • Platform builder
  • Platform manager
• How to create a BSP
  • How the platform works
    • Platform build
    • Target setup
  • Creating a package
  • Platform metadata
  • Documentation
    • Structure
    • File details
      • doc/hw/description.rst
      • doc/hw/procurement.rst
      • doc/hw/issues.rst
      • doc/sw/description.rst
      • doc/sw/procurement.rst
  • Platform support files
  • Support docker
    • deps.sh
    • build.sh
    • setup.sh
  • Manager docker
  • Builder docker
• MCU BSP

AI App

• Description
• Create LPDNN’s file tree
  • Challenge
  • Catalog
    • model.yml
    • algorithm.yml
      • Algorithm Configuration
      • Subcomponents
      • Algorithm Configuration Examples
  • Deployment
    • Algorithm Deployment specification
    • Deployment engines
      • LNE
      • TensorRT
      • NCNN
      • ONNXruntime
  • Deployment packages
• Generate an AI App
  • Prerequisites
  • Generation
    • AI App generator script
    • Bonseyes CLI
  • Development
• Execute an AI App on host device
  • Execution of FP32 AI Apps
  • Processing options
    • Input format
    • Split pre-proceesing and inference
    • Additional input
      • Extra input
      • Multiple inputs
    • Concatenation of two AI App
      • Bash
      • Python
      • C++
    • Other options
  • Quantisation workflow for LPDNN’s engines
    • TensorRT
      • FP16
      • INT8
    • ONNXruntime
      • INT8
    • NCNN
      • FP16
      • INT8
    • LNE
      • INT8
• Deploy and benchmark an AI App on target device
  • Bonseyes CLI benchmark
  • Custom deployment and benchmark
    • Host side
      • Bash
      • Python
• Upload AI App
• Create challenge-compliant interface for an AI App
  • Create header file
  • Create CLI example

LPDNN

• Description
• How to compile LPDNN’s SDK
  • Requirements
  • Getting LPDNN-SDK
  • Build for specific Target Platform
  • Run AiApp from command line
• How to add new pre- & post-processing?
  • AI App Preprocessing
    • Preprocessors
    • Driving example
      • image_preprocessor (cpp/hpp)
      • image_preprocessor_align.cpp
      • image_landmarks.cpp
      • image_based.hpp
  • AI App Postprocessing
    • Implementation
    • Output result
• How to add a new inference engine?
• How to add a new AI App class?
  • General workflow
  • Files that need to be changed
• How to create and upload a deployment package?
• How to check if an ONNX model is supported
• How to add a new operator
  • Useful links
  • General workflow
  • Files that need to be changed
  • Comparing outputs
  • Writing and running inference unit tests
• ONNX Converter Operator Definition
  • Setup
  • Writing the file
  • Available decorators
  • About ONNX Opset version