How does TAPPAS hailogallery work?

General
1

Can we get any documentation as to how exactly TAPPAS’s hailogallery works? I am trying to get multiple embeddings for a single face to work but I can’t find any documentation as to what format hailogallery expects the gallery.json file to be in.

2

Hey @M_S

I Brought this up with our R&D and they will work on it.

HailoGallery in TAPPAS Framework

Overview

HailoGallery is a component within TAPPAS that enables:

  • Saving and comparing embeddings for cross-stream object tracking
  • Managing embeddings via local JSON files
  • Object tracking using pre-saved embeddings

Key Parameters

  1. Class ID: Specifies classes to track (default: all classes)
  2. Similarity Threshold (similarity-thr): Float value (0-1) determining embedding similarity for ID matching
  3. Gallery Queue Size: Number of embeddings saved per global ID
  4. Load from Local Gallery: Option to load pre-existing embeddings from JSON file

Gallery JSON File Format

Typical structure includes:

  • IDs: Unique object identifiers
  • Embeddings: Arrays associated with each object
  • Metadata: May include class IDs, timestamps, etc.

Usage Examples

  1. Saving Faces:

    ./save_faces.sh --network scrfd_10g --clean

    Processes face images, stores embeddings in gallery JSON file

  2. Loading a Gallery:
    Enable load-local-gallery parameter and specify gallery JSON file path

This setup facilitates efficient object recognition and tracking across multiple streams.

Best Regards,
Omri

3

Is there any new information on this? I figured two months may have been enough time for some more documentation to come out but I haven’t been able to find any.

Thank you