Deep Vision is a serverless edge platform for explainable perception challenges that is focused on enabling the development and deployment of new computer vision and multi-modal spatio-temporal algorithms.
Deep Vision targets several key problems including improving analytics accuracy on spatio-temporal data, self-supervised learning, lifelong learning and creating explainable models.
We currently have 3 Deep Vision applications, RecalM, VQPy interface, and Ethosight (integration in process).
The system can be run in any environment supporting GPU and docker (NVIDIA extension as well). Make sure you have updated GPU driver on the host machine.
The detailed instructions can be found here.
Directory structure:
DeepVision
├── CONTRIBUTING.md (doc describing contibution on this repo)
├── LICENSE (A license used)
├── README.md (enty point for docs)
├── .gitignore
├── .gitmodules (configuration for git submodules)
├── .env (tracking model and corresponding accuracy score configuration)
├── docker-compose.yml (docker compose configuration)
├── Dockerfile (docker file for video sourcing and rendering)
├── mkdocs.yml (ReadTheDocs extension configuration)
├── producer.py (video reading and sourcing script)
├── requirements.txt (dependency requirements)
├── server.py (annotated video rendering server)
├── trackertotimeseries.py (timeseries labeling)
├── .github (github CI/CD workflow configuration)
├── dashboards (directory containing metrics dashboard configs for manual setting in grafana)
├── data (directory containing video source examples)
├── docs (documentation)
├── grafana (grafana provisioning and configuration)
├── recallm (system for temporal context understanding with NLP)
├── tracking (containing tacking module and submudules (MMtracking))
└── tracklet (tracking utility data structures and classes)
RecallM provides a system capable of natural language analytics by supplementing Large Language Models with an updatable, persistent memory mechanism. Click here for more details.
Detailed documentation can be found here.
Any feedback, questions, and issue reports are welcomed. Please follow Contributor Guide for more information.
For more details about this and other Cisco Research projects, please visit our home page at DeepVision Home Page