DeepStack Beta - Release Notes

DeepStack Beta is more optimized for storage, memory, compute and is more accurate.

Improvements

  • New Fast GPU Version

    We have released the GPU Version of DeepStack, accelerating responses by orders of magnitude if you have an Nvidia GPU. See Using DeepStack with NVIDIA GPU for setup instructions

  • Improved Face Recognition Engine

    The face recognition API is now powered by state-of-the-art face recognition engine.

  • Better error handling and crash recovery

    DeepStack now gracefully handles invalid images and reports errors better.

  • 50% Reduction in Install Size

    The total install size DeepStack is now half the original size.

API Changes

  • OBJECT DETECTION - MINIMUM CONFIDENCE

    By default, all objects with a confidence of 0.4 and above are detected. However, you can adjust the confidence level using the min_confidence parameter.

Breaking Changes

  • ENDPOINT ACTIVATION

    To avoid unneccesary memory usage. Only endpoints you activate are loaded.

    For example, to use any face API, you should run deepstack as below

    sudo docker run -e VISION-FACE=True -v localstorage:/datastore \
    -p 80:5000 deepquestai/deepstack
    

    In this example, you can query all face related APIs, however, you cannot query the object detection API and other endpoints.

    You can activate multiple endpoints at once as below

    sudo docker run -e VISION-FACE=True -e VISION-SCENE=True -e VISION-TRAFFIC=True \
    -v localstorage:/datastore -p 80:5000 deepquestai/deepstack
    
  • CONFIDENCE VALUES

    All endpoints now return confidence scores in range of 0 - 1. Previously, some endpoints returned confidence in range of 0 - 100

  • FACE RECOGNITION API-MINIMUM CONFIDENCE

    The min_distance parameter has been replaced by min_confidence See Face Recognition for usage instructions

  • FACE RECOGNITION API - DATA INCOMPATIBILITY

    Any faces registered with the Alpha Version is incompatible with this Version. You need to re-register previously registered faces. The new face recognition engine is stable and future releases will remain compatible with this Version.

  • FACE RECOGNITION API - RESPONSES

    Face registeration response has been changed from

    {'predictions': {'message': 'face added'}, 'success': True}
    

    To

    {'message': 'face added', 'success': True}
    

    Face delete response has been changed from

    {'success': True, 'message': 'user deleted successfully'}
    

    To

    {'success': True}
    
  • SCENE AND TRAFFIC API - RESPONSES

    Responses has been changed from

    {'success': True, 'predictions': [{'label': 'conference_room', 'confidence': 73.73981475830078}]}
    

    To

    {'success': True, 'confidence': 0.7373981, 'label': 'conference_room'}
    
  • OBJECT DECTION API - SPEED MODES

    Speed Modes have been deprecated.