5. Embedded Application Development (Runtime) Manual

This section brings you the descriptions of Embedded Application Development , runtime BPU SDK API DOC , runtime Basic Sample User Guide , runtime AI Benchmark User Guide and Tool Introduction. We provides you with a detailed introduction to the steps of application development in the D-Robotics platform, and provides APIs, data, structures, layouts, alignment rules, application development examples, tools, and more for your reference.

• 5.1. Embedded Application Development
  • 5.1.1. General Descriptions
  • 5.1.2. Create a New Project
  • 5.1.3. Implement the Project
  • 5.1.4. Compile and Run the Project
  • 5.1.5. Multi-model Control Strategy
    • 5.1.5.1. Group Control Strategy
    • 5.1.5.2. Model Preemption Control
  • 5.1.6. Suggestions on Application Optimization
  • 5.1.7. Other Dev Tools
  • 5.1.8. FAQ
    • 5.1.8.1. What is BPU memory Cache?
    • 5.1.8.2. What is the physical/virtual address in BPU memory?
    • 5.1.8.3. How to convert camera dumped NV12 images into other formats e.g. BGR etc.?
• 5.2. BPU SDK API DOC
  • 5.2.1. General Description
  • 5.2.2. Data Type and Structure
    • 5.2.2.1. Version Information Type
      • 5.2.2.1.1. HB_DNN_VERSION_MAJOR
      • 5.2.2.1.2. HB_DNN_VERSION_MINOR
      • 5.2.2.1.3. HB_DNN_VERSION_PATCH
    • 5.2.2.2. Model Type
      • 5.2.2.2.1. HB_DNN_TENSOR_MAX_DIMENSIONS
      • 5.2.2.2.2. HB_DNN_INITIALIZE_INFER_CTRL_PARAM
      • 5.2.2.2.3. hbPackedDNNHandle_t
      • 5.2.2.2.4. hbDNNHandle_t
      • 5.2.2.2.5. hbDNNTaskHandle_t
      • 5.2.2.2.6. hbDNNTaskDoneCb
      • 5.2.2.2.7. hbDNNTensorLayout
      • 5.2.2.2.8. hbDNNDataType
      • 5.2.2.2.9. hbDNNTensorShape
      • 5.2.2.2.10. hbDNNQuantiShift
      • 5.2.2.2.11. hbDNNQuantiScale
      • 5.2.2.2.12. hbDNNQuantiType
      • 5.2.2.2.13. hbDNNTensorProperties
      • 5.2.2.2.14. hbDNNTaskPriority
      • 5.2.2.2.15. hbDNNTensor
      • 5.2.2.2.16. hbDNNRoi
      • 5.2.2.2.17. hbDNNInferCtrlParam
    • 5.2.2.3. System Type
      • 5.2.2.3.1. hbBPUCore
      • 5.2.2.3.2. hbDSPCore
      • 5.2.2.3.3. hbSysMem
      • 5.2.2.3.4. hbSysMemFlushFlag
    • 5.2.2.4. Preprocess Type
      • 5.2.2.4.1. HB_DNN_INITIALIZE_RESIZE_CTRL_PARAM
      • 5.2.2.4.2. hbDNNResizeType
      • 5.2.2.4.3. hbDNNResizeCtrlParam
    • 5.2.2.5. Plugin Type
      • 5.2.2.5.1. hbDNNLayerCreator
  • 5.2.3. APIs
    • 5.2.3.1. Version Information
      • 5.2.3.1.1. hbDNNGetVersion()
    • 5.2.3.2. Model Loading and Freeing
      • 5.2.3.2.1. hbDNNInitializeFromFiles()
      • 5.2.3.2.2. hbDNNInitializeFromDDR()
      • 5.2.3.2.3. hbDNNRelease()
    • 5.2.3.3. Model Information
      • 5.2.3.3.1. hbDNNGetModelNameList()
      • 5.2.3.3.2. hbDNNGetModelHandle()
      • 5.2.3.3.3. hbDNNGetInputCount()
      • 5.2.3.3.4. hbDNNGetInputName()
      • 5.2.3.3.5. hbDNNGetInputTensorProperties()
      • 5.2.3.3.6. hbDNNGetOutputCount()
      • 5.2.3.3.7. hbDNNGetOutputName()
      • 5.2.3.3.8. hbDNNGetOutputTensorProperties()
    • 5.2.3.4. Model Inference
      • 5.2.3.4.1. hbDNNInfer()
      • 5.2.3.4.2. hbDNNRoiInfer()
      • 5.2.3.4.3. hbDNNWaitTaskDone()
      • 5.2.3.4.4. hbDNNReleaseTask()
      • 5.2.3.4.5. hbDNNSetTaskDoneCb()
    • 5.2.3.5. Memory Related Operations
      • 5.2.3.5.1. hbSysAllocMem()
      • 5.2.3.5.2. hbSysAllocCachedMem()
      • 5.2.3.5.3. hbSysFlushMem()
      • 5.2.3.5.4. hbSysFreeMem()
      • 5.2.3.5.5. hbSysWriteMem()
      • 5.2.3.5.6. hbSysReadMem()
      • 5.2.3.5.7. hbSysRegisterMem()
      • 5.2.3.5.8. hbSysUnregisterMem()
    • 5.2.3.6. Preprocess
      • 5.2.3.6.1. hbDNNResize()
    • 5.2.3.7. Plug-ins
      • 5.2.3.7.1. hbDNNRegisterLayerCreator()
      • 5.2.3.7.2. hbDNNUnregisterLayerCreator()
  • 5.2.4. Data Layout and Alignment Rules
    • 5.2.4.1. Data Layout
    • 5.2.4.2. BPU Alignment Rules and Restrictions
      • 5.2.4.2.1. Model Input
      • 5.2.4.2.2. Stride
    • 5.2.4.3. About NV12
      • 5.2.4.3.1. YUV Format
      • 5.2.4.3.2. NV12 Format
  • 5.2.5. Error Code
  • 5.2.6. Configuration Information
  • 5.2.7. Trace log configuration instructions
  • 5.2.8. Cautions for Simulator Use
• 5.3. Basic Sample User Guide
  • 5.3.1. Foreword
  • 5.3.2. Introduction of Release Package
    • 5.3.2.1. Sample Code Package
  • 5.3.3. Prerequisites
    • 5.3.3.1. Prepare the Dev Board
    • 5.3.3.2. Compilation
  • 5.3.4. How to Use
    • 5.3.4.1. Basic Samples
      • 5.3.4.1.1. quick_start
      • 5.3.4.1.2. API Tutorial
      • 5.3.4.1.3. Advanced Samples
      • 5.3.4.1.4. Miscellaneous Samples
  • 5.3.5. Helper Tools
    • 5.3.5.1. Log
      • 5.3.5.1.1. Sample Log
      • 5.3.5.1.2. The dnn Log
• 5.4. AI Benchmark User Guide
  • 5.4.1. Foreword
  • 5.4.2. About the Deliverables
    • 5.4.2.1. The Sample Code Package
    • 5.4.2.2. Sample Models
    • 5.4.2.3. Public Datasets
  • 5.4.3. Development Environment
    • 5.4.3.1. Prepare the Dev Board
    • 5.4.3.2. Compilation
  • 5.4.4. How to Use
    • 5.4.4.1. Evaluation Scripts
    • 5.4.4.2. Performance Evaluation
      • 5.4.4.2.1. How to Use Performance Evaluation Scripts
      • 5.4.4.2.2. About Command-line Parameters
      • 5.4.4.2.3. About Configuration File
    • 5.4.4.3. Model Accuracy Evaluation
      • 5.4.4.3.1. Data Pre-process
        • 5.4.4.3.1.1. PTQ Model Data Pre-process
          • 5.4.4.3.1.1.1. VOC Dataset
          • 5.4.4.3.1.1.2. COCO Dataset
          • 5.4.4.3.1.1.3. ImageNet Dataset
          • 5.4.4.3.1.1.4. Cityscapes Dataset
      • 5.4.4.3.2. Model Mounting
      • 5.4.4.3.3. Lst File Generation
      • 5.4.4.3.4. Model Inference
        • 5.4.4.3.4.1. About Command-line Parameters
        • 5.4.4.3.4.2. About Configuration File
      • 5.4.4.3.5. Model Accuracy Computing
        • 5.4.4.3.5.1. PTQ Model Accuracy Computing
          • 5.4.4.3.5.1.1. Classification Models
        • 5.4.4.3.5.2. Detection Models
          • 5.4.4.3.5.2.1. Segmentation Models
  • 5.4.5. Model Integration
    • 5.4.5.1. Add Post-processing File
    • 5.4.5.2. Add Model Execution and Configuration Files
  • 5.4.6. Helper Tools
    • 5.4.6.1. Log
      • 5.4.6.1.1. Sample Log
      • 5.4.6.1.2. dnn Log
    • 5.4.6.2. OP Time Consumption
      • 5.4.6.2.1. General Descriptions
      • 5.4.6.2.2. Sample
    • 5.4.6.3. Dump Tool
• 5.5. Tool Introduction
  • 5.5.1. hrt_bin_dump Tool Introduction
    • 5.5.1.1. Tool Introduction
    • 5.5.1.2. Description of Input Parameters
    • 5.5.1.3. Instructions for Use
      • 5.5.1.3.1. Example
  • 5.5.2. hrt_model_exec Tool Introduction
    • 5.5.2.1. Tool Introduction
    • 5.5.2.2. Description of Input Parameters
    • 5.5.2.3. Usage Instructions
      • 5.5.2.3.1. model_info
        • 5.5.2.3.1.1. Overview
        • 5.5.2.3.1.2. Example
      • 5.5.2.3.2. infer
        • 5.5.2.3.2.1. Overview
        • 5.5.2.3.2.2. Example
        • 5.5.2.3.2.3. Optional Parameters
      • 5.5.2.3.3. perf
        • 5.5.2.3.3.1. Overview
        • 5.5.2.3.3.2. Example
        • 5.5.2.3.3.3. Optional Parameters
      • 5.5.2.3.4. Multi-threaded Latency Data Description
      • 5.5.2.3.5. Multi-input Model Description
      • 5.5.2.3.6. Additional Description of Input Parameters
        • 5.5.2.3.6.1. input_file
        • 5.5.2.3.6.2. roi_infer
        • 5.5.2.3.6.3. dump_intermediate
        • 5.5.2.3.6.4. hybrid_dequantize_process
        • 5.5.2.3.6.5. profile_path
        • 5.5.2.3.6.6. enable_cls_post_process
        • 5.5.2.3.6.7. thread_num
    • 5.5.2.4. FAQ
      • 5.5.2.4.1. How are Latency and FPS data calculated?
      • 5.5.2.4.2. Why is the FPS estimated by Latency inconsistent with the FPS measured by the tool?
      • 5.5.2.4.3. How to run model with custom op?
  • 5.5.3. Common Operations
    • 5.5.3.1. Check Out Dev Board’s Image Version
    • 5.5.3.2. Check System Logs
    • 5.5.3.3. Check BPU Utilization