SPRADC9 july   2023 AM62A3 , AM62A7

 

  1.   1
  2.   Abstract
  3.   Trademarks
  4. 1Introduction
    1. 1.1 Defect Detection Demo Summary
    2. 1.2 AM62A Processor
    3. 1.3 Defect Detection Systems
    4. 1.4 Conventional Machine Vision vs Deep Learning
  5. 2Data Set Preparation
    1. 2.1 Test Samples
    2. 2.2 Data Collection
    3. 2.3 Data Annotation
    4. 2.4 Data Augmentation
  6. 3Model Selection and Training
    1. 3.1 Model Selection
    2. 3.2 Model Training and Compilation
  7. 4Application Development
    1. 4.1 System Flow
    2. 4.2 Object Tracker
    3. 4.3 Dashboard and Bounding Boxes Drawing
    4. 4.4 Physical Demo Setup
  8. 5Performance Analysis
    1. 5.1 System Accuracy
    2. 5.2 Frame Rate
    3. 5.3 Cores Utilization
    4. 5.4 Power Consumption
  9. 6Summary
  10. 7References

3.1 Model Selection

In order to execute the model on the C7x/MMA deep learning accelerator, it must be compiled/exported to a friendly format. TI’s EdgeAI-ModelZoo provides hundreds of state-of-the-art models which are converted/exported from their original training frameworks to an embedded friendly format. These models have been slightly modified to ensure the highest performance when executed on TI Deep Learning Accelerators. Some of the tasks supported by the models in the ModelZoo are image classification, object detection, semantic segmentation, human position, any several more.

The cloud-based Edge AI Studio Model Analyzer provides an easy to use Model Selection tool. It is dynamically updated to include all models supported in TI EdgeAI-ModelZoo. The tool requires no previous experience and provides an easy to use interface to enter the features required in the desired model. The “Model Selection” tool suggests several object detection models for AM62A. Selecting the final model depends on the specific application and tasks complexity. The ONR-OD-8200-yolox-nano-lite-mmdetcoco-416x416 model is chosen for the defect detection demo. This model has several appealing features including low latency with the resolution that is adequate for this application. Table 3-1 lists the important features of the model selected for the defect detection demo. The details are for the model when trained on the coco dataset with 80 classes.

Table 3-1 Highlight Features of the Model yolox-nano-lite That are Used in the Defect Detection Demo (The details are for the model when trained on the coco dataset with 80 classes,)
Model Task Resolution AP 50% Accuracy On COCO Latency/Frame (ms) DDR BW Utilization (MB/Frame)
YoloX-Nano-Lite Multi Object Detection 416x416 40.1 8.88 22