TI Deep Learning Library User Guide
TIDL: Output Post Processing in Test Application

Table of Contents

Introduction

TIDL Support Below Post Processing types in Stand alone Test application to Sanity test the model deployments. For all the post processing inFileFormat shall be 2. All the these post processing in test application are only for demo purpose not production quality.

  • postProcType - 1 : Classification Accuracy
  • postProcType - 2 : Object Detection Box drawing
  • postProcType - 3 : Pixel Level Color Blending
  • postProcType - 4 : Pixel Level MeanIOU Accuracy

Classification Accuracy

This post processing implements top-1 and top-5 accuracy computation for classification networks. The input list file shall contain the list of file with expected class index. Refer the imageNet_sample_val.txt file in the testvecs folder for reference

  • The fist number after "A:" is expected class index.
  • The Second is Top 1 Accuracy (1 means 100%) till current frame processing.
  • The Third is Top 5 Accuracy (1 means 100%) till current frame processing.
  • The Last is Predicted Class index for the current Frame
Postproc_1.PNG
Classification Accuracy

Object Detection Box drawing

This post processing Draws the detected box on the original input image. Unique color for each class. The input file shall contain the list of input images to be processed. Maximum 21 classes are supported. There is no limit on number of detected objects

Postproc_2.PNG
Object Detection BBox

Pixel Level Color Blending

This post processing colors each pixel in the input image with unique color for each class index or depth estimation.

Postproc_3.PNG
Pixel Level Semantic Segmentation

Pixel Level MeanIOU Accuracy

  • This post processing implements MeanIOU accuracy computation for Pixel level segmentation networks. The input list file shall contain the list of input images with ground truth image for each input. When Post Processing 4 is enabled, then the Pixel Level Color Blending (3)is enabled by default.