Article ID: 000093368 Content Type: Troubleshooting Last Reviewed: 02/15/2023

Different Output Results for Custom Object Detection Model Inference on CPU and MYRIAD Using OpenVINO™ 2021.4

BUILT IN - ARTICLE INTRO SECOND COMPONENT
Summary

Steps to obtain correct output results for custom object detection model on CPU and MYRIAD using OpenVINO™ 2022 releases.

Description
  1. Performed inference using custom trained model for person detection in OpenVINO™ 2021.4.
  2. Obtained huge differences in the output results for CPU as compared to MYRIAD
Resolution
  1. Install the latest OpenVINO™ release (OpenVINO™ 2022 onwards).
  2. Run the following command using Model Optimizer to convert the model:

    mo --input_model <input_model> --data_type FP16 --layout NCHW --use_new_frontend --input_shape [input_shape] --mean_values [mean_values] --scale 255

Related Products

This article applies to 1 products