BrainFrame VisionCapsule Zoo

Tests

Introduction

This repository is used to store open source VisionCapsules created by Aotu. These capsules are built using the OpenVisionCapsule format.

All of these models are compatible with and best run by BrainFrame. More information about Brainframe can be found at our website.

Repository Structure

All capsules are within the capsules/ directory. An unpackaged capsule is simply a directory with the required capsule files. For documentation on the format and how to write your own capsules, please refer to the OpenVisionCapsules docs.

Here is an example of what you might see in the capsules/ directory:

.
└── capsules
    ├── gender_classifier
    │   ├── capsule.py
    │   ├── meta.conf
    │   ├── model.LICENSE
    │   ├── model.weird_format
    │   └── README.md
    ├── person_detector
    │   ├── capsule.py
    │   ├── meta.conf
    │   ├── model.LICENSE
    │   ├── model.pb
    │   ├── README.md
    │   └── supporting_code.py
    └── vehicle_detector
        ├── capsule.py
        ├── meta.conf
        ├── model.LICENSE
        ├── README.md
        ├── openvino_model.bin
        └── openvino_model.xml

Each subdirectory, such as gender_classifier/ or person_detector/ is an unpackaged capsule that can be loaded (and automatically packaged) by BrainFrame. If you have an existing BrainFrame installation, simply copy the desired capsule directory into the BrainFrame server's capsules/ directory, and restart the server. BrainFrame will automatically package and load the capsule.

Currently all capsules in the repository were developed by Aotu, but pull requests are encouraged if you would like to contribute. The model files the capsules use are either developed by Aotu or by 3rd parties, but permissively licensed in either case. Each capsule that has a model will contain a model.LICENSE file associated with the source for that model. In addition, if the model is from a 3rd party, links to the upstream source of the model will be provided in the README.md of the capsule directory.

Git LFS

This repository uses Git LFS to store model files and other large resources. Please see the installation guide for details.

Running Tests

This repository contains a test suite that packages up and runs every capsule in the capsules directory. The tests use the vcap libraries existing test utility that fuzzes many inputs and assesses if the capsule is conforming to the standard.

Before running tests, install the necessary dependencies.

pip install -r tests/requirements.txt

Then, run the tests with pytest, at the root of the repository.

pytest .