Hello everyone,
I'm Ivo, a mechanical engineer from Croatia and the creator of BABA. I'm sharing the project with Mike's permission.
I built BABA for my own home, bringing continuous recording, local detection and an event timeline into one interface. I'd like to introduce it here and hear from people running different cameras and hardware.
Current features include:
Recording and detection run locally. Installation uses Docker Compose, with NVIDIA/TensorRT, Intel/OpenVINO and CPU variants. The CPU option is slow and mainly intended for evaluation. Performance depends on the hardware, camera count and settings.
The source code is publicly available under PolyForm Noncommercial, and personal, non-commercial use is free. It is a personal project with a recent public release, so experience across other people's hardware and cameras is still limited.
You can explore the interface in your browser using sample household data:
demo-baba.boskovic.biz
Source code, installation instructions and screenshots:
github.com
I develop the project with AI coding agents, directing the system design, testing and deployment.
If this looks useful for your setup, let me know which cameras and server/GPU you use. I can help check the requirements before you spend time installing it. Questions and feedback are welcome.
I'm Ivo, a mechanical engineer from Croatia and the creator of BABA. I'm sharing the project with Mike's permission.
I built BABA for my own home, bringing continuous recording, local detection and an event timeline into one interface. I'd like to introduce it here and hear from people running different cameras and hardware.
Current features include:
- Continuous recording without transcoding, with timeline playback and clip export.
- Person and vehicle detection across the full frame, without waiting for motion to trigger detection.
- Zones with events for entering, leaving and staying too long.
- Live camera views and an event history filtered by camera and object type.
- Notifications through email, Telegram, Slack or webhooks.
Recording and detection run locally. Installation uses Docker Compose, with NVIDIA/TensorRT, Intel/OpenVINO and CPU variants. The CPU option is slow and mainly intended for evaluation. Performance depends on the hardware, camera count and settings.
The source code is publicly available under PolyForm Noncommercial, and personal, non-commercial use is free. It is a personal project with a recent public release, so experience across other people's hardware and cameras is still limited.
You can explore the interface in your browser using sample household data:
BABA — live demo
A self-hosted NVR that understands what it sees: transformer detection on every frame, person re-identification, face recognition and scene understanding, on your own hardware. A public demo over anonymised recordings.
Source code, installation instructions and screenshots:
GitHub - Bacinac/baba: Self-hosted NVR that understands what it sees: transformer detection on every frame, person re-identification, face recognition and scene understanding.
Self-hosted NVR that understands what it sees: transformer detection on every frame, person re-identification, face recognition and scene understanding. - Bacinac/baba
github.com
I develop the project with AI coding agents, directing the system design, testing and deployment.
If this looks useful for your setup, let me know which cameras and server/GPU you use. I can help check the requirements before you spend time installing it. Questions and feedback are welcome.