mirror of
https://github.com/blakeblackshear/frigate.git
synced 2026-05-07 05:55:27 +03:00
NVR with realtime local object detection for IP cameras
aicameragoogle-coralhome-assistanthome-automationhomeautomationmqttnvrobject-detectionrealtimertsptensorflow
71060805f0
Three independent ORT/glibc leak vectors identified and fixed: 1. **ORT CPU BFC arena** (`enable_cpu_mem_arena=False` for all sessions) ORT's default CPU arena pools host-side GPU↔CPU staging buffers indefinitely. Disabling it across every InferenceSession (detection + embedding) stops hundreds-of-MB/h RSS growth seen on systems with CUDA EP sessions. 2. **ORT memory-pattern cache** (`enable_mem_pattern=False` for variable-length models) For embedding models with variable-length inputs (Jina v1/v2, PaddleOCR), ORT allocates one mmap-backed execution plan per unique sequence length and never frees them. Disabling the pattern cache stops this unbounded anon-mmap growth. Fixed-size models (YOLO) keep `enable_mem_pattern=True` to preserve buffer aliasing and avoid CUDA graph capture failures. 3. **mallopt(M_ARENA_MAX)** called from `EmbeddingProcess.run()` The forkserver start method exec()s a fresh Python interpreter that does not inherit Docker env vars, so `MALLOC_ARENA_MAX` set in docker-compose never reaches the child. Calling `mallopt(-8, os.cpu_count())` from `run()` caps glibc malloc arenas in the child process. Additional improvements: - `compute_cuda_mem_limit()`: dynamically caps the ORT CUDA EP BFC arena for embedding sessions to min(model_size × 7, 80% VRAM); prevents OOM on multi-model systems while leaving headroom for detection sessions. - CUDA graph capture is now wrapped in try/except so models with CPU-only ops (e.g. attention, NMS) fall back to ONNXModelRunner instead of crashing. - `ONNXModelRunner.has_variable_length_inputs()`: centralises the Jina/PaddleOCR detection logic to keep SessionOptions creation consistent. - 17 regression-guard unit tests in `frigate/test/test_detection_runners.py` that will fail if any of these three fixes is accidentally reverted. Fixes: #23007 Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> |
||
|---|---|---|
| .cspell | ||
| .cursor/rules | ||
| .devcontainer | ||
| .github | ||
| .vscode | ||
| config | ||
| docker | ||
| docs | ||
| frigate | ||
| migrations | ||
| notebooks | ||
| testing-scripts | ||
| web | ||
| .dockerignore | ||
| .gitignore | ||
| .pylintrc | ||
| audio-labelmap.txt | ||
| CODEOWNERS | ||
| CONTRIBUTING.md | ||
| cspell.json | ||
| docker-compose.yml | ||
| generate_config_translations.py | ||
| labelmap.txt | ||
| LICENSE | ||
| Makefile | ||
| netlify.toml | ||
| package-lock.json | ||
| pyproject.toml | ||
| README_CN.md | ||
| README.md | ||
| TRADEMARK.md | ||
Frigate NVR™ - Realtime Object Detection for IP Cameras
EnglishA complete and local NVR designed for Home Assistant with AI object detection. Uses OpenCV and Tensorflow to perform realtime object detection locally for IP cameras.
Use of a GPU or AI accelerator is highly recommended. AI accelerators will outperform even the best CPUs with very little overhead. See Frigate's supported object detectors.
- Tight integration with Home Assistant via a custom component
- Designed to minimize resource use and maximize performance by only looking for objects when and where it is necessary
- Leverages multiprocessing heavily with an emphasis on realtime over processing every frame
- Uses a very low overhead motion detection to determine where to run object detection
- Object detection with TensorFlow runs in separate processes for maximum FPS
- Communicates over MQTT for easy integration into other systems
- Records video with retention settings based on detected objects
- 24/7 recording
- Re-streaming via RTSP to reduce the number of connections to your camera
- WebRTC & MSE support for low-latency live view
Documentation
View the documentation at https://docs.frigate.video
Donations
If you would like to make a donation to support development, please use Github Sponsors.
License
This project is licensed under the MIT License.
- Code: The source code, configuration files, and documentation in this repository are available under the MIT License. You are free to use, modify, and distribute the code as long as you include the original copyright notice.
- Trademarks: The "Frigate" name, the "Frigate NVR" brand, and the Frigate logo are trademarks of Frigate, Inc. and are not covered by the MIT License.
Please see our Trademark Policy for details on acceptable use of our brand assets.
Screenshots
Live dashboard
Streamlined review workflow
Multi-camera scrubbing
Built-in mask and zone editor
Translations
We use Weblate to support language translations. Contributions are always welcome.
Copyright © 2026 Frigate, Inc.