* feat(deepx): add DEEPX NPU detector and runtime integration.
* feat(deepx): enforce model_format requirement when ppu is enabled and add integrity checks for driver installation
* Update frigate/detectors/plugins/deepx.py
Public method lacks docstring
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* Refactor DEEPX detector tests, support SSD and DAMO-YOLO
* feat(deepx): add anchor-free output decoding and corresponding tests
* Add tests and updates for DEEPX detector and refactor DEEPX accelerator code structure.
* fix: enhance model type validation and update documentation for DEEPX detector
* fix: add support for customizable score and NMS thresholds
* refactor: infer YOLO layout from the model, drop per-detector options and the dxrtd placeholder
* fix: keep only the anchor-free PPU verdict, re-read anchor-based each frame
* Update latency data for DEEPX NPU
* Expanding PPU support for DEEPX and set yolo-generic as default.
* enhance scale count resolution logic
* Extend PPU layout handling and YOLOX support to DEEPX detector
* fix: assume the largest PPU anchor table when the .dxnn has no layout
* Improve PPU decoding and introduce strides handling
* Improve PPU scope and fix box format mismatch
* Fix unnamed node issue that breaks traversal
* fix: update object detection model type description to remove outdated architecture
---------
Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>
* don't display audio transcription provider message as health notice
* show remote provider for audio transcription in health pane
* adjust trigger and notifications messages to be consistent with the rest of the settings UI
* disable save buttons when there are no changes in config editor
* fix audio manager crash when a camera is added at runtime
The audio processor and the camera maintainer both poll the same `add` config update on their own one second timers, and the maintainer is what creates `camera_metrics[name]`. When the audio processor got there first it looked the new camera up before that entry existed, and the `KeyError` took down the whole `frigate.audio_manager` process. Whether it happens depends purely on which poll fires first, so cloning a camera from the UI fails or succeeds at random. `spawn_if_needed` now skips a camera whose metrics aren't there yet and picks it up on the next poll, the same way it already waits on a late ffmpeg update.
`AudioEventMaintainer` holds the `CameraMetrics` object now instead of indexing the manager dict on every audio chunk, which drops the IPC round trips and means a removed camera can't `KeyError` out of `detect_audio` after the maintainer pops the entry. The audio process is also registered with the watchdog, since a crash there previously left audio detection dead for every camera until a full restart, and it now receives the shared `DataProcessorMetrics` so `AudioTranscriptionRealTimeProcessor` gets the same type as the other real time processors.
* fix stationary max_frames dropping other tracked objects
When `max_frames` was set for a label, deregistering one object rebuilt norfair's list with a filter that kept an object only if it was both not the target and already on its way out, so every other healthy object of that label was dropped along with it. Any car leaving the frame took the rest of the cars with it and they came back as new tracked objects a few frames later. The filter now removes only the target, and objects that are expiring are still reaped by norfair on the next update.
* fix test
* fix skip_motion_threshold permanently disabling motion detection
The skip check returned before the two `accumulateWeighted` calls at the end of `detect`, so a skipped frame never made it into the background and setting `calibrating` there only picked a faster alpha for calls that never ran. `avg_frame` starts as an all zero image and a normally lit scene differs from black across nearly the whole frame, so the cameras I tested measure 0.84 to 0.98 against it. Any `skip_motion_threshold` below that number skips the first frame, leaves the background black, and skips every frame after it. Motion detection is dead for that camera until the setting is removed or Frigate restarts, with no motion boxes, no motion recordings, and no regions for the tracker since the detector stays calibrating.
Startup isn't the only way in. `update_mask` zeroes the background on any motion config change, and once a camera has calibrated the first IR switch or PTZ move freezes the background on the old scene, so it can't transition to the new one, which is the case the option exists for. The frame is now blended in before the early return at the same 0.2 alpha the calibrating path uses elsewhere, so a large scene change is still suppressed while the background catches up, about a second on a 5 fps camera, and then motion comes back.
* dump ffmpeg logs on every restart
The record watchdog restarted ffmpeg without flushing its `LogPipe`, so a camera whose recording segments went stale never showed a single line of ffmpeg output. The dump now happens in `start_or_restart_ffmpeg` right after the stop, which covers the stale record path, the record crash path, and the audio restart. `reset_capture_thread` and the audio `log_and_restart` fallback keep their own dumps since both pass `ffmpeg_process=None`.
* dump ffmpeg logs once per restart
The audio restart path dumped the log pipe itself before calling the helper, so the restart dump printed a second "last 100 lines" heading over an already drained deque and split the tail that `stop_ffmpeg` flushed into its own section. The heading is now only printed when there's something under it, and the audio path leaves the dump to the restart so each failure produces one section.
* keep all logpipe dumps consistent
* check for a valid frame before using its shape
With the camera offline, no preview frame, and `camera-error.jpg` missing, `latest_frame` read `frame.shape` before its `frame is None` check, so it raised `AttributeError` and answered 500 with a traceback instead of the intended "Unable to get valid frame". The check now runs first.
* fix the has_clip self-heal for events with no recordings
`vod_event` looked for a `(body, 404)` tuple, but `vod_ts` returns a `JSONResponse`, so the check never matched and an old event whose recordings are gone kept offering a clip that can't play. It now checks the response status code.
* return 403 for a snapshot or thumbnail on another camera
The broad `except Exception` handlers in `event_snapshot` and `event_thumbnail` caught the `HTTPException` from `require_camera_access`, so a restricted user asking for another camera's snapshot got a 404 instead of a 403, and for an object still being tracked the snapshot was rendered before the check ran. Both endpoints now look up the event and check access in their own block, the way the other endpoints do, so a denial propagates.
* find DST transitions to the second
`get_dst_transitions` probed the offset once every 24 hours from the start time and reported a change at the first probe after it, up to a day late, so events, review items and recordings near a transition were grouped into days with the old offset. A transition after the last daily probe wasn't found at all. The end of the range is probed too now, and a probe that sees the offset change bisects the interval to the second of the transition.
* don't run page shortcuts for keys a dialog already handled
Radix dismisses a dialog on Escape from a capture-phase keydown listener and calls `preventDefault()` without stopping propagation, so `useKeyboardListener` still ran the page's Escape shortcut: cancelling the delete dialog in the face library or a classification model also cleared the whole selection. Keys another shortcut hook handled still get through, since their listener order changes with every render.
* fix train image filtering for a class with a dash
The backend writes a class with a `-` as `_` in train file names, since it splits those names on `-`, while a dataset folder keeps the dash. Filtering the Train grid by `half-open` compared it with `half_open` and hid every attempt. Both sides are normalized the same way now.
* don't edit a chat message while a reply streams
The edit button stayed active while a reply streamed. `submitConversation` returns early while loading, but the message bubble still closed its editor, so the edit was silently lost. The edit button is hidden while a reply streams, and an editor that's already open keeps its draft with send disabled until the reply ends.
* fix restart failing under non-root
restart_frigate() called psutil.Process(1).terminate() to signal s6-svscan, but s6-svscan runs as root while frigate runs as uid 1000, so the call raised AccessDenied. That exception escaped every caller: the UI restart button dropped its websocket client, MQTT restart and Save & Restart just logged and did nothing, and the watchdog crashed its own monitoring thread on a dead detector. This catches AccessDenied and falls through to the existing SIGINT branch, which exits the process for s6 to restart it.
* show runtime overrides in the settings form
The settings form read a camera section's saved config value, but its dependent warnings (audio transcription requiring audio detection, snapshots requiring detect, etc.) read the live config instead. A runtime toggle from the live view, MQTT, or an active profile can turn a section off without touching yaml, and that override persists across restarts, so the Enable switch showed on while the warning said the feature wasn't enabled. This adds an "Overridden (Live)" badge to any field whose live value differs from what's saved, and swaps the affected warnings to runtime-specific wording when a runtime override is the actual cause instead of the config.
* fix mobile overflowing icons in system due to new health pane
* fix genai settings keeping a stale model and dropping roles after save
Switching a GenAI entry's provider left the previous provider's model selected, so saving wrote a model the new provider doesn't serve. llama.cpp can't find that model in `/v1/models`, so the backend reported every capability as false for the entry, and once the save refetched `genai/models` the roles widget stripped `transcribe` from the form on its own. The section showed unsaved changes right after saving, and saving again would have dropped the role. Switching provider now clears the model, and the roles widget only strips a role for a model or provider picked in the form, since the entry-level capability flags only describe the saved model. A selected role stays visible when the provider can't confirm it, so it can still be switched off. The llama.cpp model list also no longer repeats a model whose alias matches its id, which is what `--alias` produces.
* close onvif sessions on shutdown
`OnvifController.close()` only stopped its event loop, so the aiohttp sessions each `ONVIFCamera` holds and the `_poll_config_updates` task were left to be garbage collected during interpreter shutdown, when their warnings can no longer be logged. Every restart ended with a run of `Unclosed client session` and `Task was destroyed but it is pending!` logging errors, which only became visible once restart started exiting the process itself under non-root. `close()` now closes each camera's client and cancels the tasks on the loop before stopping it.
* fixes
* fixes
Validate config exists 0 regardless of if the config is valid or not.
This makes it not very useful for CI
Tiny fix to bail non-zero if the config is invalid
* Add support for running transcription with GenAI
* Improve audio joining
* Fix GenAI model capability reporting
* Support language correctly
* Migrate existing users to keep english selected
* Fix models
* Fix tests
* Fix accepted null model
* Handle slwo providers
* Implement annotated frames mode for GenAI reviews to improve models with lacking temporal understanding
* Updates
* Improve debug sharing
* Do not number objects
* Fix assumptions
* Remove unhelpful content
* Improve object data sent as part of prompt
* Cleanup ollama dumbness
* Bind db
* Fixes
* Cleanup
* improve zone renaming
Zone rename now saves as one JSON body via config/set, moving required_zones and profile overrides instead of leaving stale references
* fixes
* allow users to select live streaming technology
* fix webrtc being downgraded to mse on load
`useUserPersistence` seeds state with the default and loads asynchronously, so the first render always used `mse` instead of the saved choice, and `useWebRTCGloballyAvailable` reports `checking` until the probe settles and re-enters that state on every consumer mount, so a saved `webrtc` was rewritten to `mse` even after the probe had already passed. On Safari the MSE player then timed out and latched the jsmpeg fallback. A pending probe now counts as available, the player waits on `autoLive` until the stored preferences load, and `handleError` gates on the mode in use since the fallback flag no longer implies webrtc is untried. A rejected IndexedDB read also resolves `loaded` now, so a blocked store can't leave the player waiting forever.
* add support for configurable ICE servers in WebRTC player
* add mic error state, fix dialog overwriting saved choice and dashboard ignoring stream
* tweaks
fix duplicated styles, move fonts to src/assets/fonts for vite to bundle (nginx already rewrites correctly), and fix fast refresh for PreviewController, auth context/provider, and statusbar context
* update date-fns, i18next, react-dropzone, react-markdown and @types/node
react-i18next 17 peers `i18next >= 26.2.0`, so the two move together. react-day-picker already depends on date-fns 4, so date-fns now dedupes to a single copy. react-dropzone 20 declares `node >=22` in `engines`, but npm only warns on Node 20 and nothing in the build needs Node 22.
* update js-yaml, @hookform/resolvers and prettier
js-yaml 5 has no default export, so `DictAsYamlField` imports `dump`, `load` and `YAMLException` by name. @hookform/resolvers 5 types `zodResolver` with the schema's input and output types separately, and fields with defaults are optional on input, so the zone and classification model forms pass both types to `useForm`. zod's range now starts at 3.25, which resolvers 5 requires.
* format with prettier 3.9
* remove unused web test deps
Nothing runs vitest. The CI step that called `npm run test` is commented out, `web/__test__/` was deleted in https://github.com/blakeblackshear/frigate/pull/8983 so `setupFiles` points at a missing file, and there are no unit tests, so `npx vitest run` only picks up the Playwright specs and fails. jsdom, `@testing-library/jest-dom`, msw and fake-indexeddb were only there for vitest.
* update contributing docs
* remove unused deps
konva 10.5 removed the private `Node._lastPos`, so `PolygonCanvas` now reads the dragged point from `getAbsolutePosition()`, which returns the position konva just applied. monaco-yaml 5.5 takes formatter options instead of a boolean for `format`. The radix packages move together so every shared primitive stays a single copy under the `react-slot` and `compose-refs` overrides.
* fix recordings unavailable endpoint when no params are provided
we already import the datetime class directly, so those attribute lookups raised AttributeError and the request returned 500
* reject JWTs whose role is no longer in the config
`/auth` trusted the role inside the JWT and re-signed it on refresh without checking the config, so a user whose restricted role was deleted kept a session carrying a role that isn't in `auth.roles`. The media, clip, recording, export and go2rtc checks treat a missing role the same as a role with no camera list, so that session could open every camera. `/auth` now returns 401 for a token whose role isn't configured, which sends the user back through login, and login already falls back to `viewer` for a role that's gone.
* delete the deleted camera group's layout, not the open one's
Deleting a camera group removed the layout of the group being viewed, because the dialog's delete was bound to `${activeGroup}-draggable-layout`. It also cleared the saved group, and both ran before the `config/set` request whether or not it succeeded, so deleting one group while viewing another lost the open group's layout and left the deleted group's layout behind. The deleted group's own layout is now removed after a successful save, and the saved group is only cleared when it was the open group.
* don't block API when querying PTZ info
camera_ptz_info is async but waited on the ONVIF controller's future with future.result(), blocking the API event loop for as long as a slow or unreachable ONVIF camera took to answer (including reconnect attempts), so every other async request stalled with it. Await the future with asyncio.wrap_future instead. The coroutine runs on the controller's own loop and thread, so this cannot deadlock.
* for custom exports, only allow admin users to add to existing cases
follows the existing convention where attaching an export to an existing case is admin-only on `POST /export/{camera}/...` and `POST /exports/batch`
* drop pending edits for a camera or profile that no longer exists
* match cached preview frames to their camera exactly
https://github.com/blakeblackshear/frigate/pull/22594 added a trailing `-` to the `preview_{camera}` prefix so `camera` stopped matching `camera2`'s frames, but camera names can contain `-`, so `front` still matched `front-door`'s. After a restart, `front`'s preview recorder deleted this hour's frames of a matching camera that sorted before it and added the timestamps of one that sorted after it, so ffmpeg was asked for files that don't exist and that hour's preview was lost. The offline fallback for `latest.jpg` could also return `front-door`'s frame for `front`, even to a user without access to `front-door`, and a short export could take its fallback thumbnail from the other camera. These now compare the full camera name taken from the file name.
* migrate web to eslint 10 flat config
ESLint 10 dropped `.eslintrc` support, so `.eslintrc.cjs` is replaced with `eslint.config.js` and the lint scripts no longer pass `--ext` or `--ignore-path`. typescript-eslint moves to 8, react-hooks to 7, and react-refresh to 0.5, and the unused jest and vitest-globals plugins are removed. Lint behaves as it did before: catch variables aren't checked, unused disable directives aren't reported, and rules newly added to the recommended sets are off until the code passes them. typescript-eslint 8 flags constants used only in `typeof`, so those are now exported, or replaced with a union type where the export would trip react-refresh.
* fix lint findings from the eslint 10 recommended rules
Remove the rule overrides from the flat config migration and fix what they were hiding. Unused catch bindings are dropped, 20 disable directives that suppressed nothing are removed (react-hooks 5.2 and 7.1.1 report identical exhaustive-deps findings with inline config ignored), dead initial values are dropped, short-circuit calls become if statements or optional calls, rethrown errors pass `cause`, and the disabled "No recordings" tooltip in `ReviewTimeline` is removed along with its memo and the `getRecordingAvailability` prop. The 3 react-refresh warnings for files that export contexts or classes are left for a later refactor.
* update docusaurus to 3.10.2
* bump @types/node to 25.9.6 and ES2022
target ES2022, which already includes ES2020 and ES2021.String, so the lib list was also trimmed
* bump vite to 8.3.0 and vitest to 4.1.11
Swap `@vitejs/plugin-react-swc` for `@vitejs/plugin-react` and add `esbuild` as a devDependency, since `vite-plugin-monaco-editor` requires it and Vite 8 no longer ships it. `keepNames` moves to `build.rolldownOptions.output` because Vite 8 ignores the `esbuild` block. Rolldown's minifier now writes the preload helper's base path as a template literal instead of a double-quoted string, so the nginx `sub_filter` for `return"/BASE_PATH/"` stopped matching and lazy-loaded chunks and their CSS were requested from a literal `/BASE_PATH/` under Home Assistant ingress. The rule now matches the backtick form.
* bump apexcharts to 7.3.0 and react-apexcharts to 2.1.1
Under Vite 8, a default import from a CommonJS package resolves to its whole `module.exports` when `package.json` has `"type": "module"`, so react-apexcharts 1.4.1 handed React an object and every chart crashed. 2.x ships an ESM build. apexcharts 7 no longer sets `window.ApexCharts`, so the chart components now import it for `ApexCharts.exec`, and `ApexAxisChartSeries` is derived in `types/graph.ts` because it's no longer a global type.
* remove unused immer dep
* remove unused cython pin from tensorrt requirements
The pin was added alongside tensorrt 8.5.3 and cuda-python 11.8, which needed Cython to build, and both have since been dropped from this file. Nothing imports Cython at runtime, and `pip3 wheel` runs with build isolation, so any source build gets its own build dependencies. The pin only installed an unused Cython wheel into the TensorRT image.
* require node 20.19 for docs
types-peewee 4.0 types model fields precisely, so 17 `type: ignore` comments and 2 `cast(str, ...)` calls are no longer needed. Its stubs type `.namedtuples()` and `.dicts()` queries as returning model instances, so the review cleanup reads namedtuple fields by name and the storage usage query casts its dict rows. `start_time` is declared `DateTimeField` but stores unix timestamps, so two reads cast it like `debug_replay.py` already does. `Export` gets an annotation for the `export_case_id` attribute peewee adds at runtime.
description:Visible on the System Metrics page in the Web UI. Please include the full version including the build identifier (eg. 0.18.0-beta1, 0.18.0-8b72c7a, etc.)
placeholder:"0.18.0-beta1"
description:Visible on the System Metrics page in the Web UI. Please include the full version including the build identifier (eg. 0.19.0-beta1, 0.19.0-8b72c7a, etc.)
Use this form to submit a reproducible bug in Frigate or Frigate's UI.
**⚠️ If you are running a beta version (0.18.0-beta or similar), please use the [Beta Support template](https://github.com/blakeblackshear/frigate/discussions/new?category=beta-support) instead.**
If you are running on Proxmox, please see the [Proxmox FAQ](https://github.com/blakeblackshear/frigate/discussions/23916) and reproduce the issue on a standard Docker install first (bare metal, or a VM running plain Debian/Ubuntu) before submitting here.
**⚠️ If you are running a beta version (0.19.0-beta or similar), please use the [Beta Support template](https://github.com/blakeblackshear/frigate/discussions/new?category=beta-support) instead.**
Before submitting your bug report, please ask the AI with the "Ask AI" button on the [official documentation site][ai] about your issue, [search the discussions][discussions], look at recent open and closed [pull requests][prs], read the [official Frigate documentation][docs], and read the [Frigate FAQ][faq] pinned at the Discussion page to see if your bug has already been fixed by the developers or reported by the community.
download:If no custom model path or URL is provided, the Hailo detector automatically downloads the default model (YOLOv6n) from the Hailo Model Zoo on first startup based on the detected hardware. Once cached under `/config/model_cache/hailo`, the model works fully offline.
download:If no custom model path or URL is provided, the Hailo detector automatically downloads the default model (YOLOv6n) from the Hailo Model Zoo on first startup, choosing the build that matches the attached device. Once cached under `/config/model_cache`, the model works fully offline.
ui:|-
Navigate to **Settings > System > Detection models** and select **Hailo** from the **Hardware** dropdown. Then, on the same model, open the **Custom Model** tab and configure the model settings:
@@ -60,7 +60,7 @@ hailo8l:
yaml:|-
models:
- devices:
- hailo8l:PCIe
- hailo:PCIe
width: 320
height: 320
input_tensor: nhwc
@@ -101,7 +101,7 @@ hailo8l:
yaml:|-
models:
- devices:
- hailo8l:PCIe
- hailo:PCIe
width: 300
height: 300
input_tensor: nhwc
@@ -824,24 +824,6 @@ cpu:
models:
- devices:
- cpu:3
deepstack:
title:DeepStack / CodeProject.AI
models:
- key:yolo
label:YOLO
recommended:true
download:This detector runs object detection over the network against a CodeProject.AI or DeepStack server, so no model is downloaded into Frigate itself. Visit the [CodeProject.AI official website](https://www.codeproject.com/Articles/5322557/CodeProject-AI-Server-AI-the-easy-way) to download and install the AI server on your preferred device (e.g. Raspberry Pi, Nvidia Jetson, or other compatible hardware) before configuring the detector.
ui:|-
Navigate to **Settings > System > Detection models** and add a model. The CodeProject.AI server is not reported by the hardware probe, so set `devices` to `deepstack:http://<your_codeproject_ai_server_ip>:<port>/v1/vision/detection` in YAML.
# ├── ssdlite_mobilenet.dfp (a file ending with .dfp)
# └── ssdlite_mobilenet_post.onnx (optional; only if the model includes a cropped post-processing network)
deepx:
title:DEEPX NPU
models:
- key:yolo
label:YOLO
recommended:true
download:Nomodel is bundled with Frigate. Download a pre-compiled YOLO `.dxnn` model from the [DEEPX ModelZoo](https://developer.deepx.ai/modelzoo) or compile your own with DX-COM, then bind-mount it into the container and point the model's `path` at it. The recommended model is `yolox-s_640x640_ppu.dxnn`. Its Post-Processing Unit (PPU) compile moves candidate selection onto the NPU, which makes it the fastest ModelZoo model measured through Frigate (about 13 ms on a DX-M1). The output layout is read from the compiled model, so anchor-based, anchor-free, NMS-in-head and PPU models (anchor-based or anchor-free) all need no extra configuration; prefer a `PPU` variant whenever the ModelZoo offers one. PPU models must be compiled with DX-COM 2.4.0 or later, which writes the head layout Frigate reads into the file.
ui:|-
Navigate to **Settings > System > Detection models** and select **DEEPX NPU** from the **Hardware** dropdown. Then, on the same model, open the **Custom Model** tab and configure:
download:Nomodel is bundled with Frigate. Download a pre-compiled YOLOX `.dxnn` model from the [DEEPX ModelZoo](https://developer.deepx.ai/modelzoo), then bind-mount it into the container and point the model's `path` at it. The `_ppu` variant is faster and also works with the `yolo-generic` model type; the plain export needs `yolox` so its raw head is decoded.
ui:|-
Navigate to **Settings > System > Detection models** and select **DEEPX NPU** from the **Hardware** dropdown. Then, on the same model, open the **Custom Model** tab and configure:
import AnalyticsFields from "@site/src/components/AnalyticsFields";
import NavPath from "@site/src/components/NavPath";
Frigate can send one anonymous usage report a day. The reports show the maintainers which hardware to support, which features people use, and how releases perform. Sharing is off until you turn it on.
## Turning it on
Enable **Share anonymous analytics** at <NavPath path="Settings > System > Telemetry" />, or set it in your config:
```yaml
telemetry:
analytics:true
```
The same page has a **Preview the report** button that shows exactly what the next report contains.
## How it's sent
- Once a day, as a JSON POST to `https://analytics.frigate.video/report`
- The server looks up your country and region from your IP address and never stores the address
- Raw reports are kept for 60 days; only aggregate totals are published
- A random install ID, stored in `/config/.analytics.json`, keeps your install from being counted twice. Turning sharing off deletes it
## What's never sent
- Camera, zone, group, profile, or user names
- Object labels, face names, or license plate text
- IP addresses, hostnames, URLs, or stream paths
- Credentials or API keys
- Events, recordings, or anything from them
## Every field
Fields marked public appear in the published totals. The machine-readable schema is [frigate-analytics-schema.json](pathname:///frigate-analytics-schema.json).
@@ -204,7 +204,7 @@ Frequently-heard labels like `speech` can generate a lot of events, and each eve
### Audio Transcription
Frigate supports fully local audio transcription using either `sherpa-onnx` or OpenAI's open-source Whisper models via `faster-whisper`. The goal of this feature is to support Semantic Search for `speech` audio events. Frigate is not intended to act as a continuous, fully-automatic speech transcription service. Automatically transcribing all speech (or queuing many audio events for transcription) requires substantial CPU (or GPU) resources and is impractical on most systems. For this reason, transcriptions for events are initiated manually from the UI or the API rather than being run continuously in the background.
Frigate supports fully local audio transcription using either `sherpa-onnx` or OpenAI's open-source Whisper models via `faster-whisper`, and can alternatively offload transcription to a [GenAI provider](#genai-provider). The goal of this feature is to support Semantic Search for `speech` audio events. Frigate is not intended to act as a continuous, fully-automatic speech transcription service. Automatically transcribing all speech (or queuing many audio events for transcription) requires substantial CPU (or GPU) resources and is impractical on most systems. For this reason, transcriptions for events are initiated manually from the UI or the API rather than being run continuously in the background.
:::info
@@ -224,6 +224,7 @@ To enable transcription, configure it globally and optionally disable for specif
- Set **Audio transcription model or GenAI provider name** to `whisper` for Frigate's built-in local models, or to the name of a GenAI provider
- Set **Transcription device** to the desired device
- Set **Model size** to the desired size
@@ -235,6 +236,7 @@ To enable transcription, configure it globally and optionally disable for specif
```yaml
audio_transcription:
enabled:True
model:whisper
device:...
model_size:...
```
@@ -263,20 +265,88 @@ The optional config parameters that can be set at the global level include:
- **`enabled`**: Enable or disable the audio transcription feature.
- Default: `False`
- It is recommended to only configure the features at the global level, and enable it at the individual camera level.
- **`model`**: The transcription backend.
- Default: `whisper`
-`whisper` uses Frigate's built-in local models, described by `device` and `model_size` below.
- Any other value must name a key in your `genai` config whose entry has `transcribe` in its `roles`. See [GenAI Provider](#genai-provider).
- **`device`**: Device to use to run transcription and translation models.
- Default: `CPU`
- This can be `CPU` or `GPU`. The `sherpa-onnx` models are lightweight and run on the CPU only. The `whisper` models can run on GPU but are only supported on CUDA hardware.
- Ignored when `model` names a GenAI provider.
- **`model_size`**: The size of the model used for live transcription.
- Default: `small`
- This can be `small` or `large`. The `small` setting uses `sherpa-onnx` models that are fast, lightweight, and always run on the CPU but are not as accurate as the `whisper` model.
- This config option applies to **live transcription only**. Recorded `speech` events will always use a different `whisper` model (and can be accelerated for CUDA hardware if available with `device: GPU`).
- **`language`**: Defines the language used by `whisper` to translate `speech` audio events (and live audio only if using the `large` model).
- Default: `en`
-You must use a valid [language code](https://github.com/openai/whisper/blob/main/whisper/tokenizer.py#L10).
- This config option applies to **live transcription only**. With `model: whisper`, recorded `speech` events always use a different `whisper` model (and can be accelerated for CUDA hardware if available with `device: GPU`).
- Ignored when `model` names a GenAI provider.
- **`language`**: Defines the language used to transcribe and translate `speech` audio events (and live audio only if using the `large` model or a GenAI provider).
-Default: `auto`
-`auto` lets the model detect the language itself, which most models do well. Set an explicit language only if detection is picking the wrong one.
- Otherwise you must use a valid [language code](https://github.com/openai/whisper/blob/main/whisper/tokenizer.py#L10).
- Transcriptions for `speech` events are translated.
- Live audio is translated only if you are using the `large` model. The `small``sherpa-onnx` model is English-only.
The only field that is valid at the camera level is `enabled`.
The only field that is valid at the camera level is `enabled`. In particular `model` is global only: the transcription backend is a process-wide resource shared by every camera.
#### GenAI Provider
Frigate can send audio to a GenAI provider for transcription when that provider has the `transcribe` role. This is useful if you already run a GenAI provider, or if you do not have the CPU/GPU headroom for a local whisper model. Supported providers are **OpenAI**, **Azure OpenAI**, **Gemini**, and **llama.cpp** with an audio-capable model (a dedicated ASR model such as Qwen3-ASR, or a general multimodal model that accepts audio). Ollama is not supported as it has no audio input.
To use a GenAI provider for audio transcription:
1. Configure a GenAI provider with `transcribe` in its `roles`.
2. Set the audio transcription model to that GenAI config key (e.g. `whisper_cloud`).
<ConfigTabs>
<TabItem value="ui">
Navigate to <NavPath path="Settings > Enrichments > Audio transcription" />.
| **Audio transcription model or GenAI provider name** | Set to the GenAI config key (e.g. `whisper_cloud`) to use a configured GenAI provider for transcription |
The GenAI provider must also be configured with the `transcribe` role under <NavPath path="Settings > Enrichments > Generative AI" />.
</TabItem>
<TabItem value="yaml">
```yaml
genai:
whisper_cloud:
provider:openai
api_key:your-api-key
model:gpt-transcribe
roles:
- transcribe
audio_transcription:
enabled:True
model:whisper_cloud
language:en
```
</TabItem>
</ConfigTabs>
:::warning
**Give `transcribe` its own `genai` entry.** A `genai` entry has a single `model` string that is shared by every role it holds, so `roles: [descriptions, transcribe]` would send the same model name to both the chat endpoint and the transcription endpoint. Transcription models and chat models are almost never the same model, so define a dedicated entry as shown above.
:::
:::warning
**Live transcription against a metered provider is billed continuously.** In live mode Frigate uploads an overlapping ~2 second window of audio roughly once per second, per camera, for as long as audio stays above that camera's `audio.min_volume`. Windows below that threshold are never uploaded, which is what keeps a quiet camera near zero requests, but a camera pointed at a busy street will keep sending.
Three things keep this opt-in: `transcribe` is not one of the default roles, live transcription is off by default, and the volume gate suppresses silence. Transcription of recorded `speech` events is unaffected - it remains a manual, one-request-per-event action.
:::
`device` and `model_size` have no effect on this path and no local model is ever downloaded.
`language` defaults to `auto`, which sends no language hint and lets the model detect it. Most audio models detect language well, so leave it on `auto` unless detection is picking the wrong one.
When set explicitly, it is sent as the transcription endpoint's native `language` parameter for OpenAI, Azure, and llama.cpp, and as part of the prompt for Gemini. This matters for dedicated ASR models such as Qwen3-ASR: they read the prompt as contextual biasing rather than as an instruction, so a language named in the prompt is ignored, while the endpoint parameter is honored.
#### Live transcription
@@ -292,6 +362,8 @@ Results can be error-prone due to a number of factors, including:
For speech sources close to the camera with minimal background noise, use the `small` model.
A [GenAI provider](#genai-provider) is generally the most accurate option for live transcription, at the cost of a network round trip per window. That round trip has to stay under about a second to keep up with the audio; if it does not, Frigate drops the oldest buffered audio rather than letting the backlog grow.
If you have CUDA hardware, you can experiment with the `large``whisper` model on GPU. Performance is not quite as fast as the `sherpa-onnx``small` model, but live transcription is far more accurate. Using the `large` model with CPU will likely be too slow for real-time transcription.
#### Transcription and translation of `speech` audio events
@@ -308,7 +380,7 @@ Only one `speech` event may be transcribed at a time. Frigate does not automatic
:::
Recorded `speech` events will always use a `whisper` model, regardless of the `model_size` config setting. Without a supported Nvidia GPU, generating transcriptions for longer `speech` events may take a fair amount of time, so be patient.
With `model: whisper`, recorded `speech` events always use a `whisper` model, regardless of the `model_size` config setting. Without a supported Nvidia GPU, generating transcriptions for longer `speech` events may take a fair amount of time, so be patient. With a [GenAI provider](#genai-provider), the recorded clip is sent to the provider instead and no local model is used.
The examples on this page all use `my_provider`, but the name is arbitrary and is only used to reference the provider elsewhere in the config (for example, `semantic_search.model`).
Each provider handles one or more **roles**: `chat`, `descriptions`, and`embeddings`. A provider handles all three by default, and each role may be assigned to exactly one provider. Define a single provider if you want it to do everything, or split the roles across several providers using the `roles` option.
Each provider handles one or more **roles**: `chat`, `descriptions`, `embeddings`, and `transcribe`. A provider handles the first three by default; `transcribe` must always be listed explicitly, and is not available on Ollama, which has no audio input. Each role may be assigned to exactly one provider. Define a single provider if you want it to do everything, or split the roles across several providers using the `roles` option.
If the provider you choose requires an API key, you may either directly paste it in your configuration, or store it in an environment variable prefixed with `FRIGATE_`.
@@ -63,11 +63,11 @@ Running Generative AI models on CPU is not recommended, as high inference times
You must use a vision-capable model with Frigate. The following models are recommended for local deployment of the `descriptions` and `chat` roles:
| `qwen3-vl` | Strong visual and situational understanding, enhanced ability to identify smaller objects and interactions with object. |
| `qwen3.6`/`qwen3.8` | Strong situational understanding, but missing DeepStack from qwen3-vl leading to worse performance for identifying objects in people's hand and other small details. |
| `gemma4` | Strong situational understanding, sometimes resorts to more vague terms like 'interacts' instead of assigning a specific action. |
| Model | Review [frame mode](/configuration/genai/genai_review#frame-mode) | Notes |
| `qwen3-vl` |`frames` | Strong visual and situational understanding, enhanced ability to identify smaller objects and interactions with object. Follows a sequence of frames on its own. |
| `qwen3.6`/`qwen3.8` |`frames` | Strong situational understanding, but missing DeepStack from qwen3-vl leading to worse performance for identifying objects in people's hand and other small details. |
| `gemma4` |`annotated_frames` | Strong situational understanding, sometimes resorts to more vague terms like 'interacts' instead of assigning a specific action. Loses track of activity that repeats or reverses, so it benefits from annotated frames. |
#### Embedding models
@@ -77,6 +77,17 @@ The `embeddings` role needs a different kind of model. Text queries are matched
| `qwen3-vl-embedding` | Multimodal embeddings for [Semantic Search](/configuration/semantic_search#genai-provider). Must be served by llama.cpp started with `--embeddings` and `--mmproj`. |
#### Transcription models
The `transcribe` role needs a model that accepts audio input. A text-only or vision-only model cannot serve this role. The following are recommended for local deployment of the `transcribe` role:
| `qwen3-asr` | Dedicated speech recognition model covering 30 languages, and the better choice for transcription quality. It only transcribes, so it cannot be shared with the `descriptions` or `chat` roles. |
| `gemma4` | General multimodal model that accepts audio as well as images, so one served model can cover `transcribe` alongside the other roles. Transcript quality is below `qwen3-asr`, particularly on noisy audio. |
Both must be served by llama.cpp started with the matching audio `--mmproj`. llama.cpp only reports audio support when an audio projector is loaded. Without it Frigate sees the model as text-only and the `transcribe` role is unavailable in the UI. Frigate transcribes through the server's `/v1/audio/transcriptions` route, which llama.cpp serves for any audio-capable model.
:::info
Each model is available in multiple parameter sizes (3b, 4b, 8b, etc.). Larger sizes are more capable of complex tasks and understanding of situations, but requires more memory and computational resources. It is recommended to try multiple models and experiment to see which performs best.
Review items are sent to the model as a sequence of still frames. Some models follow that sequence well on their own; others lose track of activity that repeats or reverses, and describe a single trip when the subject actually made several. The `frame_mode` option controls how those frames are presented.
- `frames` (default): the prompt followed by the frames, exactly as earlier versions of Frigate sent them.
- `annotated_frames`: each frame is preceded by its frame number and elapsed time, along with notes describing what the object tracker recorded at that moment, such as an object being first detected, starting to move, turning around, stopping, or no longer being detected.
The notes come from tracking data rather than from the images, so they describe activity the model may not have picked up on its own. In testing with a person carrying three waste bins to the curb one at a time, `gemma4` described a single trip on every attempt with `frames`, and consistently described multiple trips with `annotated_frames`. Models that already handle these sequences well, such as the `qwen3-vl` family, gain little and should stay on `frames`.
Annotated mode also caps the number of frames, since the notes already establish the order of events and extra near-duplicate frames tend to crowd out the middle of a clip. Longer review items are sampled more sparsely as a result, and typically use fewer tokens than `frames` mode for the same item.
:::note
Annotated mode needs tracking data for the review item. If none is available, Frigate falls back to sending plain frames for that item.
:::
<ConfigTabs>
<TabItem value="ui">
Navigate to <NavPath path="Settings > Global configuration > Review" />.
- Set **GenAI config > Frame mode** to the desired mode (e.g., `annotated_frames`)
</TabItem>
<TabItem value="yaml">
```yaml {4}
review:
genai:
enabled: true
frame_mode: annotated_frames
```
</TabItem>
</ConfigTabs>
### Response Style
Different models respond to the built-in prompt with very different writing styles: some produce natural narration while others sound short and mechanical. The `response_style` option selects a writing style preset that rewords the prompt's instructions for the user-facing fields (the title, short summary, and scene description). Presets replace those instructions rather than adding extra ones, so the model never receives competing style directions.
@@ -28,6 +28,24 @@ WebRTC may use an external STUN server for NAT traversal. MSE and HLS streaming
:::
### Selecting a streaming technology
Frigate [defaults to MSE](#why-does-frigate-prefer-mse-over-webrtc-for-live-view) for restreamed cameras by design. To use WebRTC, select it explicitly from a camera's single-camera Live view settings (the settings menu in the camera's Live view header on desktop, or the settings drawer on mobile). Three related controls work together:
- **Stream**: _what_ to play. This lists the [streams you've configured](#setting-streams-for-live-ui) (for example `Main Stream` and `Sub Stream`).
- **Force low-bandwidth mode**: a switch that always plays Frigate's built-in low-bandwidth feed (the stream assigned the `detect` role, using JSMpeg) instead of the selected stream. It works anywhere without go2rtc and is useful on slow or metered connections. While it is enabled, the stream and streaming technology selectors are disabled; your stream and technology choices are restored when you turn it off.
- **Streaming Technology**: _how_ to play the selected stream, listing **MSE** and **WebRTC**. It is only shown for a restreamed stream.
- The choices are saved **per device, per camera** in your browser's local storage.
- **WebRTC is only selectable when it can actually work for that stream.** When it can't, the option is shown disabled with the reason inline, and a more detailed reason (the failing codecs, or why the connectivity check failed) is logged to your browser's console. Common reasons:
- **Not configured**: no `candidates` or `ice_servers` are set under `go2rtc.webrtc` (see [WebRTC extra configuration](#webrtc-extra-configuration)).
- **Could not connect**: e.g. port `8555` isn't reachable, or a STUN/TURN server is misconfigured. Frigate runs a one-time WebRTC connectivity check when the Live view opens; the option may briefly show as "checking" while it runs.
- **Unsupported video codec**: the stream's video codec can't be played over WebRTC in your browser, most commonly H.265/HEVC in Firefox or Edge.
- **Unsupported audio codec**: WebRTC needs opus or G.711 audio, so a stream whose playback audio is only AAC (without an added opus/G.711 track) can't carry audio over WebRTC. See [Audio Support](#audio-support) for how to add one.
- **Unsupported browser**: the browser doesn't support WebRTC.
When WebRTC isn't available, Frigate automatically uses MSE (or falls back to JSMpeg), so live view keeps working regardless of the selection.
### Camera Settings Recommendations
If you are using go2rtc, you should adjust the following settings in your camera's firmware for the best experience with Live view:
@@ -157,6 +175,17 @@ WebRTC works by creating a TCP or UDP connection on port `8555`. However, it req
- stun:8555
```
- The web UI uses the STUN and TURN servers in `ice_servers` and falls back to Google's public STUN server when none are set:
```yaml title="config.yml"
go2rtc:
webrtc:
ice_servers:
- urls: [turn:turn.example.com:3478]
username: frigate
credential: password
```
- For access through Tailscale, the Frigate system's Tailscale IP must be added as a WebRTC candidate. Tailscale IPs all start with `100.`, and are reserved within the `100.64.0.0/10` CIDR block.
- Note that some browsers may not support H.265 (HEVC). You can check your browser's current version for H.265 compatibility [here](https://github.com/AlexxIT/go2rtc?tab=readme-ov-file#codecs-madness).
@@ -206,6 +235,8 @@ For devices that support two way talk, Frigate can be configured to use the feat
- Ensure you access Frigate via https (may require [opening port 8971](/frigate/installation/#ports)).
- For the Home Assistant Frigate card, [follow the docs](http://card.camera/#/usage/2-way-audio) for the correct source.
The two-way talk control in the single-camera Live view is only enabled when WebRTC is available; if WebRTC isn't configured or can't connect, the control is shown disabled.
To use the Reolink Doorbell with two way talk, you should use the [recommended Reolink configuration](/configuration/camera_specific#reolink-cameras)
As a starting point to check compatibility for your camera, view the list of cameras supported for two-way talk on the [go2rtc repository](https://github.com/AlexxIT/go2rtc?tab=readme-ov-file#two-way-audio). For cameras in the category `ONVIF Profile T`, you can use the [ONVIF Conformant Products Database](https://www.onvif.org/conformant-products/)'s FeatureList to check for the presence of `AudioOutput`. A camera that supports `ONVIF Profile T` _usually_ supports this, but due to inconsistent support, a camera that explicitly lists this feature may still not work. If no entry for your camera exists on the database, it is recommended not to buy it or to consult with the manufacturer's support on the feature availability.
@@ -313,8 +313,16 @@ To remove root from the container entirely, add Docker's `user:`:
```yaml
user:"1000:1000"# NOT compatible with PUID/PGID, see the run modes table
tmpfs:
- /tmp:size=256m
- /tmp/cache:size=1000000000
- /run:exec,nosuid,nodev,mode=0755,uid=1000,gid=1000,size=16m# uid must match user:
```
`/run` has to be owned by that uid as well. s6 writes its runtime state there before anything else starts, and with no root in the container a root-owned `/run` stops it during init with `cannot create /run/test of writability`. Keep `uid` and `gid` in the tmpfs options matching `user:`, and don't carry that pair back into the default mode, where a root-owned `/run` is what keeps the unprivileged services out of s6's runtime state.
This only bites once root is genuinely gone. s6's init helper is setuid, so `user:` on its own still lets init regain root and correct `/run` itself. The `no-new-privileges:true` above is what blocks that, which is also what makes the `/run` ownership mandatory. Dropping it would hide the problem by handing init root again.
Two things change, and the first one will break a working install if you skip it. The startup device grants can't run, because there is no root left to run them, so every device you pass stops working until you grant that uid access yourself with `group_add:` or a udev rule; see [Manual setup](#manual-setup). Expect this to surface as a driver error rather than a permission error, like `No VA display found` from VAAPI. And every service then runs as that one uid, so go2rtc no longer gets its own restricted user. `/config` and `/media/frigate` have to be owned by that uid already, since Frigate never adjusts ownership in this mode. Switching an existing install over also leaves `/config/go2rtc_homekit.yml` owned by the go2rtc user, which this mode can't write; `chown` it to your uid or HomeKit pairing changes stop persisting. Frigate warns and starts either way.
This mode can also take `cap_drop: [ALL]`, which the default mode cannot: starting as root needs `CAP_CHOWN` for the ownership sweep, `CAP_SETUID` and `CAP_SETGID` to drop to the runtime user, and `CAP_FOWNER` for the device grants.
@@ -22,8 +22,9 @@ Frigate supports multiple different detectors that work on different types of ha
**Most Hardware**
- [Coral EdgeTPU](#edge-tpu-detector): The Google Coral EdgeTPU is available in USB, Mini PCIe, and m.2 formats allowing for a wide range of compatibility with devices.
- [Hailo](#hailo-8): The Hailo8 and Hailo8L AI Acceleration module is available in m.2 format with a HAT for RPi devices, offering a wide range of compatibility with devices.
- [Hailo](#hailo): The Hailo-8, Hailo-8L and Hailo-8R AI Acceleration modules are available in m.2 format with a HAT for RPi devices, offering a wide range of compatibility with devices.
- <CommunityBadge /> [MemryX](#memryx-mx3): The MX3 Acceleration module is available in m.2 format, offering broad compatibility across various platforms.
- <CommunityBadge /> [DEEPX](#deepx-npu): The DEEPX NPU is available in m.2 format and as a HAT+ for the Raspberry Pi 5, offering broad compatibility across various platforms.
**AMD**
@@ -285,9 +286,9 @@ models:
---
## Hailo-8
## Hailo
This detector is available for use with both Hailo-8 and Hailo-8L AI Acceleration Modules. The integration automatically detects your hardware architecture via the Hailo CLI and selects the appropriate default model if no custom model is specified.
This detector is available for use with the Hailo-8, Hailo-8L and Hailo-8R AI Acceleration Modules. The integration identifies which of them is attached and selects the matching default model if no custom model is specified.
See the [installation docs](../frigate/installation.md#hailo-8) for information on configuring the Hailo hardware.
@@ -308,11 +309,11 @@ The HailoRT runtime is not part of the Frigate image. It is downloaded and insta
When configuring the Hailo detector, you have two options to specify the model: a local **path** or a **URL**.
If both are provided, the detector will first check for the model at the given local path. If the file is not found, it will download the model from the specified URL. The model file is cached under `/config/model_cache/hailo`.
For additional ready-to-use models, please visit: https://github.com/hailo-ai/hailo_model_zoo
Hailo8 supports all models in the Hailo Model Zoo that include HailoRT post-processing. You're welcome to choose any of these pre-configured models for your implementation.
Hailo supports all models in the Hailo Model Zoo that include HailoRT post-processing. You're welcome to choose any of these pre-configured models for your implementation.
> **Note:**
> The config.path parameter can accept either a local file path or a URL ending with .hef. When provided, the detector will first check if the path is a local file path. If the file exists locally, it will use it directly. If the file is not found locally or if a URL was provided, it will attempt to download the model from the specified URL.
@@ -362,7 +363,7 @@ Intel NPUs cannot be used under Home Assistant OS, which does not include the NP
:::warning
The network-based detectors (Deepstack and the Apple Silicon client) are being reworked. Their extra options no longer have a place in the config, so only the endpoint carried in the device string is honored right now: Deepstack ignores `api_key` and `api_timeout`, and the Apple Silicon client ignores`request_timeout_ms` and `linger_ms`. Anything else is dropped when your config is migrated.
The Apple Silicon detector client is being reworked. Its extra options no longer have a place in the config, so only the endpoint carried in the device string is honored right now, and`request_timeout_ms` and `linger_ms` are ignored. Anything else is dropped when your config is migrated.
:::
@@ -540,30 +541,6 @@ A TensorFlow Lite model is provided in the container at `/cpu_model.tflite` and
When using CPU detectors, you can add one CPU detector per camera. Adding more detectors than the number of cameras should not improve performance.
## Deepstack / CodeProject.AI Server Detector
:::warning
The network-based detectors (Deepstack and the Apple Silicon client) are being reworked. Their extra options no longer have a place in the config, so only the endpoint carried in the device string is honored right now: Deepstack ignores `api_key` and `api_timeout`, and the Apple Silicon client ignores `request_timeout_ms` and `linger_ms`. Anything else is dropped when your config is migrated.
:::
The Deepstack / CodeProject.AI Server detector for Frigate allows you to integrate Deepstack and CodeProject.AI object detection capabilities into Frigate. CodeProject.AI and DeepStack are open-source AI platforms that can be run on various devices such as the Raspberry Pi, Nvidia Jetson, and other compatible hardware. It is important to note that the integration is performed over the network, so the inference times may not be as fast as native Frigate detectors, but it still provides an efficient and reliable solution for object detection and tracking.
### Setup {#setup-deepstack}
To get started with CodeProject.AI, visit their [official website](https://www.codeproject.com/Articles/5322557/CodeProject-AI-Server-AI-the-easy-way) to follow the instructions to download and install the AI server on your preferred device. Detailed setup instructions for CodeProject.AI are outside the scope of the Frigate documentation.
To integrate CodeProject.AI into Frigate, configure the detector as follows:
Replace `<your_codeproject_ai_server_ip>` and `<port>` with the IP address and port of your CodeProject.AI server.
To verify that the integration is working correctly, start Frigate and observe the logs for any error messages related to CodeProject.AI. Additionally, you can check the Frigate web interface to see if the objects detected by CodeProject.AI are being displayed and tracked properly.
# Community Supported Detectors
## MemryX MX3
@@ -636,6 +613,63 @@ For detailed instructions on compiling models, refer to the [MemryX Compiler](ht
---
## DEEPX NPU
This detector is available for use with the DEEPX NPU, both the DX-M1 M.2 module and the DX-M1M on the Sixfab AI HAT+ for the Raspberry Pi 5. The configuration below applies unchanged to either form factor. DEEPX NPU support in Frigate is developed and maintained by [Sixfab](https://sixfab.com).
See the [installation docs](../frigate/installation.md#deepx-npu) for information on installing the DEEPX kernel driver and runtime on the host and passing the NPU through to the container.
To run a model on a DEEPX NPU, list a `deepx` device on that model.
:::info
The DX-RT Python bindings are not part of the Frigate image. They are downloaded and installed into `/config/.local` the first time a DEEPX device is configured, verified against pinned checksums, and updated automatically when a Frigate release pins a new version. If the container has no internet access, see [Detector runtimes](/frigate/network_requirements#detector-runtimes) for how to provide the files yourself.
Frigate does not bundle a model for this detector. Models must be compiled to DEEPX's `.dxnn` format. Two model types are supported:
- `yolo-generic` for YOLO object detection models, the recommended default. The detector reads the model's output layout from the compiled file, so anchor-based, anchor-free and NMS-in-head models all work with the same configuration, as do models compiled with DEEPX's Post-Processing Unit (PPU) support.
- `yolox` for YOLOX models compiled without PPU support, whose raw head needs Frigate's YOLOX decoder. A YOLOX model compiled with PPU support works under either `yolox` or `yolo-generic`.
The quickest way to get one is the [DEEPX ModelZoo](https://developer.deepx.ai/modelzoo), which publishes pre-compiled `.dxnn` files for a range of YOLO object detection models. Download the `.dxnn`, bind-mount it into the container, and point the model's `path` at it. Alternatively, compile your own model with the DX-COM compiler. The recommended starting point is `yolox-s_640x640_ppu.dxnn`, the fastest ModelZoo model measured through Frigate:
For PPU models, use a `.dxnn` compiled with DX-COM 2.4.0 or later. Frigate reads the PPU head layout the compiler writes into the file and refuses to load a PPU model without it.
`model_type` must be set to `yolo-generic` or `yolox` to match the model; `yolo-generic` is the recommended default unless the model is a raw YOLOX export. Frigate defaults it to `ssd`, which this detector does not support, so the detector refuses to start on a model that leaves it unset.
`width` and `height` must match the resolution the model was compiled for. Quantization parameters are baked into the `.dxnn` file at compile time, so no normalization is applied on the host and Frigate's default `input_tensor`, `input_pixel_format`, and `input_dtype` values do not need to be overridden.
A DEEPX device is `PCIe:<index>`, as reported on the detector settings page. The NPU daemon multiplexes across processes, so the same device may be listed more than once to run additional inference processes against it:
```yaml
models:
- devices:
- deepx:PCIe:0
- deepx:PCIe:0
```
#### Label maps
The object detection models in the DEEPX ModelZoo are trained on the standard 80-class COCO label set, so `labelmap_path` must be set to `/labelmap/coco-80.txt`. Frigate's default label map uses an extended 91-class COCO scheme, and leaving it in place will cause detections to be reported as the wrong object type. For `yolo-generic` models the label map is also what the detector uses to tell the output layout, so a label map with the wrong number of classes is reported as an error at startup.
---
## NVidia TensorRT Detector
Nvidia Jetson devices may be used for object detection using the TensorRT libraries. Due to the size of the additional libraries, this detector is only provided in images with the `-tensorrt-jp6` tag suffix, e.g. `ghcr.io/blakeblackshear/frigate:stable-tensorrt-jp6`. This detector is designed to work with Yolo models for object detection.
@@ -204,11 +204,20 @@ Light guidelines and advice:
npm run lint
```
-Add to unit tests and ensure they pass. As much as possible, you should strive to _increase_ test coverage whenever making changes. This will help ensure features do not accidentally become broken in the future.
- If you run into error messages like "TypeError: Cannot read properties of undefined (reading 'context')" when running tests, this may be due to these issues (https://github.com/vitest-dev/vitest/issues/1910, https://github.com/vitest-dev/vitest/issues/1652) in vitest, but I haven't been able to resolve them.
-Ensure the backend [unit tests](#unit-tests) pass. Your PR cannot be merged unless tests pass.
```shell
python3 -u -m unittest
```
- Ensure the end-to-end tests pass. They run in Playwright against a production build with mocked API data, so they don't need a running Frigate instance. Add or update tests in `web/e2e/specs/` when you change UI behavior.
```console
npm run test
# First-time setup
npx playwright install chromium
# Build the app and run all tests
npm run e2e:build && npm run e2e
```
- Test in different browsers. Firefox, Chrome, and Safari all have different quirks that make them unique targets to interact with.
@@ -54,7 +54,7 @@ Frigate supports multiple different detectors that work on different types of ha
**Most Hardware**
- [Hailo](#hailo-8): The Hailo8 and Hailo8L AI Acceleration module is available in m.2 format with a HAT for RPi devices offering a wide range of compatibility with devices.
- [Hailo](#hailo-8): The Hailo-8, Hailo-8L and Hailo-8R AI Acceleration modules are available in m.2 format with a HAT for RPi devices offering a wide range of compatibility with devices.
- [Supports many model architectures](../../configuration/object_detectors#configuration-hailo)
- Runs best with tiny or small size models
@@ -65,6 +65,11 @@ Frigate supports multiple different detectors that work on different types of ha
- [Supports many model architectures](../../configuration/object_detectors#memryx-mx3)
- Runs best with tiny, small, or medium-size models
- <CommunityBadge /> [DEEPX](#deepx-npu): The DEEPX NPU is available in m.2 format and as a HAT+ for the Raspberry Pi 5, allowing for a wide range of compatibility with devices.
- [Supports YOLO model architectures](../../configuration/object_detectors#deepx-npu)
- Runs best with tiny or small size models
- Runs efficiently on low power hardware
**AMD**
- [ROCm](#rocm---amd-gpu): ROCm can run on AMD Discrete GPUs to provide efficient object detection
@@ -111,12 +116,13 @@ Frigate supports multiple different detectors that work on different types of ha
### Hailo-8
Frigate supports both the Hailo-8 and Hailo-8L AI Acceleration Modules on compatible hardware platforms, including the Raspberry Pi 5 with the PCIe hat from the AI kit. The Hailo detector integration in Frigate automatically identifies your hardware type and selects the appropriate default model when a custom model isn’t provided.
Frigate supports the Hailo-8, Hailo-8L and Hailo-8R AI Acceleration Modules on compatible hardware platforms, including the Raspberry Pi 5 with the PCIe hat from the AI kit. The Hailo detector integration in Frigate identifies which of them is attached and selects the matching default model when a custom model isn’t provided.
**Default Model Configuration:**
- **Hailo-8L:** Default model is **YOLOv6n**.
- **Hailo-8:** Default model is **YOLOv6n**.
- **Hailo-8L:** Default model is **YOLOv6n**, compiled for the Hailo-8L.
- **Hailo-8:** Default model is **YOLOv6n**, compiled for the Hailo-8.
- **Hailo-8R:** Default model is the **Hailo-8** build of **YOLOv6n**, since the Hailo Model Zoo publishes no Hailo-8R build.
In real-world deployments, even with multiple cameras running concurrently, Frigate has demonstrated consistent performance. Testing on x86 platforms, with dual PCIe lanes, yields further improvements in FPS, throughput, and latency compared to the Raspberry Pi setup.
@@ -256,6 +262,32 @@ The MX3 is a pipelined architecture, where the maximum frames per second support
Inference speeds may vary depending on the host platform. The above data was measured on an **Intel 13700 CPU**. Platforms like Raspberry Pi, Orange Pi, and other ARM-based SBCs have different levels of processing capability, which may limit total FPS.
### DEEPX NPU
Frigate supports the DEEPX NPU in both of its form factors: the **DX-M1** M.2 module, which works on x86 (Intel/AMD) and ARM-based SBCs such as the Raspberry Pi 5, and the **DX-M1M** on the [Sixfab AI HAT+](https://docs.sixfab.com/docs/ai-hat-plus-raspberry-pi-5-quickstart) for the Raspberry Pi 5. Both use the same driver and runtime, so the configuration is identical for either one. DEEPX NPU support in Frigate is developed and maintained by [Sixfab](https://sixfab.com).
The DEEPX driver and runtime run on the Docker host rather than inside the Frigate container and must be installed before the NPU can be used. See the [installation docs](installation.md#deepx-npu) for the setup steps and [the detector docs](/configuration/object_detectors#deepx-npu) for the configuration.
Frigate does not bundle a model for this detector. Models use DEEPX's `.dxnn` format, and pre-compiled YOLO models can be downloaded from the [DEEPX ModelZoo](https://developer.deepx.ai/modelzoo). Prefer a model with a `_ppu` suffix whenever one is available for the architecture you want: these run part of the post-processing on the NPU itself and are considerably faster, roughly 2.5x for the same architecture and input size. **YOLOX-S with PPU is the recommended starting point.**
Inference times for a few recommended models, measured through Frigate's own stats on a DX-M1:
Other ModelZoo YOLO variants are also supported but have not been measured. Inference speeds vary with the host platform, so a slower host such as a Raspberry Pi 5 will report higher times than those above.
:::note
A few ModelZoo models can not be used with Frigate: SSD models (they are trained on Pascal VOC, so their labels do not match Frigate's), DAMO-YOLO models, face and pose models, and the PPU builds of YOLOv7.
:::
### Nvidia Jetson
Jetson devices are supported via the TensorRT or ONNX detectors when running Jetpack 6. It will [make use of the Jetson's hardware media engine](/configuration/hardware_acceleration_video#nvidia-jetson) when configured with the [appropriate presets](/configuration/ffmpeg_presets#hwaccel-presets), and will make use of the Jetson's GPU and DLA for object detection when configured with the [TensorRT detector](/configuration/object_detectors#nvidia-tensorrt-detector).
@@ -122,7 +122,7 @@ Additionally, the USB Coral draws a considerable amount of power. If using any o
### Hailo-8
The Hailo-8 and Hailo-8L AI accelerators are available in both M.2 and HAT form factors for the Raspberry Pi. The M.2 version typically connects to a carrier board for PCIe, which then interfaces with the Raspberry Pi 5 as part of the AI Kit. The HAT version can be mounted directly onto compatible Raspberry Pi models. Both form factors have been successfully tested on x86 platforms as well, making them versatile options for various computing environments.
The Hailo-8, Hailo-8L and Hailo-8R AI accelerators are available in both M.2 and HAT form factors for the Raspberry Pi. The M.2 version typically connects to a carrier board for PCIe, which then interfaces with the Raspberry Pi 5 as part of the AI Kit. The HAT version can be mounted directly onto compatible Raspberry Pi models. Both form factors have been successfully tested on x86 platforms as well, making them versatile options for various computing environments.
The HailoRT runtime is not part of the Frigate image; Frigate downloads and installs it at first start once a Hailo detector is configured. Containers without internet access can provide the files themselves, see [Detector runtimes](/frigate/network_requirements#detector-runtimes).
@@ -300,7 +300,7 @@ If you are using `docker run`, add this option to your command `--device /dev/ha
#### Configuration
Finally, configure [hardware object detection](/configuration/object_detectors#hailo-8) to complete the setup.
Finally, configure [hardware object detection](/configuration/object_detectors#hailo) to complete the setup.
### MemryX MX3
@@ -381,6 +381,99 @@ If you can't use Docker Compose, you can run the container with something simila
Finally, configure [hardware object detection](/configuration/object_detectors#memryx-mx3) to complete the setup.
### DEEPX NPU
The DEEPX NPU is available in two form factors, and Frigate supports both:
- **DX-M1** in the M.2 2280 form factor (like an NVMe SSD), for x86 (Intel/AMD) PCs, the Raspberry Pi 5, and other ARM SBCs with an exposed PCIe M.2 slot.
- **DX-M1M** on the [Sixfab AI HAT+](https://docs.sixfab.com/docs/ai-hat-plus-raspberry-pi-5-quickstart), a HAT+ board that connects to the Raspberry Pi 5 over PCIe Gen 3 x1.
Both present the NPU through the same PCIe driver and DX-RT runtime, so the setup below and the detector configuration are identical for either one. Nothing needs to change when moving between them.
DEEPX NPU support in Frigate is developed and maintained by [Sixfab](https://sixfab.com).
#### Versions
A DEEPX install has several separately versioned pieces, and they all have to agree. The driver, the runtime, and the daemon live on the Docker host; Frigate itself carries only the Python bindings, which it downloads on first start:
| Component | Version | Installed on | Installed by |
| NPU firmware | `v2.7.4` | The module | Flashed from the host |
| DX-RT bindings | `v3.4.0` | Frigate | Downloaded at first start |
:::warning
A version mismatch does not produce a startup error. It typically shows up as inference requests that are accepted but never return a result, so detections simply stop appearing while Frigate looks healthy. If that happens after a Frigate upgrade, check every version in the table before anything else.
:::
The installation script installs the DX-RT runtime on the host and enables `dxrt.service`, so the daemon starts at boot and any other program on the host can share the NPU with Frigate. Check the firmware version with `dxrt-cli --status` and update the module if it does not match the table above.
#### Installation
The DEEPX kernel driver must be installed on the host rather than in the container, because containers share the host kernel and cannot load kernel modules. Installing it creates the `/dev/dxrt*` device nodes that are passed through to Frigate. The same script installs the DX-RT runtime and enables `dxrt.service`, the daemon that owns the NPU and hands work to it on behalf of Frigate and anything else on the host.
1. Copy or download [this script](https://github.com/blakeblackshear/frigate/blob/dev/docker/deepx/user_installation.sh).
2. Ensure it has execution permissions with `sudo chmod +x user_installation.sh`
3. Run the script with `./user_installation.sh`
4. **Restart your computer** to complete driver installation.
Confirm the NPU is visible before continuing:
```bash
ls /dev/dxrt*
```
Then confirm the daemon is running and listening in `/run/dxrt`:
```bash
systemctl is-active dxrt.service
ls /run/dxrt/
```
#### Setup
To set up Frigate, follow the default installation instructions, for example: `ghcr.io/blakeblackshear/frigate:stable`
#### Docker configuration
Frigate needs the NPU device node and the directory holding the daemon's socket:
```yaml
services:
frigate:
devices:
- /dev/dxrt0:/dev/dxrt0
volumes:
- /run/dxrt:/run/dxrt
```
If you can't use Docker Compose, add `--device /dev/dxrt0:/dev/dxrt0 -v /run/dxrt:/run/dxrt` to your `docker run` command.
Add one `--device` per NPU, contiguously from `/dev/dxrt0`, since the client stops enumerating at the first gap.
The installation script configures `dxrt.service` to place its socket in `/run/dxrt` through a systemd drop-in. Mounting the directory rather than the socket file means the container sees the new socket after `dxrt.service` is restarted, rather than holding on to a deleted one.
`dxrtd` listens on an abstract socket as well, but that one does not cross into a container, so Frigate names the filesystem socket through `DXRT_DYNAMIC_IPC_ENDPOINT` on your behalf. Set that variable on the container yourself only if the daemon listens somewhere else, which means you also set it for `dxrtd` through its own systemd drop-in. The script writes `/etc/systemd/system/dxrt.service.d/frigate.conf` for exactly that, and has `dxrt.service` link the socket to `/tmp/dxrt_dynamic_ipc.sock` when it starts, so the host's own `dxrt-cli` and `dxtop` keep finding it at the default path they fall back to.
:::note
The DX-RT client exits when `dxrt.service` stops, so restart the Frigate container after restarting `dxrt.service`.
:::
The device node is needed as well as the socket, because the client opens the NPU directly even though the daemon arbitrates access. Without it, inference fails with `Device not found`.
`/dev/shm` does not need sharing.
The DX-RT python bindings are not shipped in the Frigate image. Frigate downloads them on first start when a DEEPX detector is configured, and caches them under `/config`.
#### Configuration
Finally, configure [hardware object detection](/configuration/object_detectors#deepx-npu) to complete the setup.
### Rockchip platform
Make sure that you use a linux distribution that comes with the rockchip BSP kernel 5.10 or 6.1 and necessary drivers (especially rkvdec2 and rknpu). To check, enter the following commands:
| [AXERA AXEngine](/configuration/object_detectors) | Detection model | HuggingFace |
:::note
@@ -60,16 +60,16 @@ The default CPU, EdgeTPU, and OpenVINO object detection models are bundled into
The SDKs for a few hardware detectors are not shipped in the Frigate image. They are downloaded the first time that detector is configured, verified against checksums pinned in the Frigate release, and installed into the Frigate user's home directory (`/config/.local` by default). Once installed they are not downloaded again until a Frigate release pins a new version.
If the container cannot reach GitHub, provide the files yourself:
1. Download the files for your architecture on a machine with internet access.
2. Place them, with exactly the file names listed above, in `/config/model_cache/runtimes/<detector>/`, where `<detector>` is the detector `type` from your config (`hailo8l`, `memryx`, or `axengine`).
2. Place them, with exactly the file names listed above, in `/config/model_cache/runtimes/<detector>/`, where `<detector>` is the detector named in your config's `devices` (`hailo`, `memryx`, or `axengine`).
3. Start Frigate. Files whose checksum matches are installed without any download; a file with the wrong checksum is discarded and downloaded again, so a failed startup log names the file to replace.
The `GITHUB_ENDPOINT` mirror variable below applies to these downloads as well.
@@ -134,6 +134,15 @@ telemetry:
version_check:false
```
### Anonymous Analytics
If [anonymous analytics](/configuration/advanced/analytics) sharing is turned on, Frigate sends one report a day to `https://analytics.frigate.video`. It's off by default, so no outbound connection happens unless you enable it:
```yaml
telemetry:
analytics:true
```
### Push Notifications
When [notifications](/configuration/notifications) are enabled and users have registered for push notifications in the web UI, Frigate sends push messages through the browser vendor's push service (e.g., Google FCM, Mozilla autopush). This requires internet access from the Frigate server to these push endpoints.
@@ -142,16 +151,12 @@ When [notifications](/configuration/notifications) are enabled and users have re
If an [MQTT broker](/integrations/mqtt) is configured, Frigate maintains a connection to the broker's host and port. This is typically a local network connection, but will require internet if you use a cloud-hosted MQTT broker.
### DeepStack / CodeProject.AI
When using the [DeepStack detector plugin](/configuration/object_detectors), Frigate sends images to the configured API endpoint for inference. This is typically local but depends on where the service is hosted.
## WebRTC (STUN)
For [WebRTC live streaming](/configuration/live), Frigate uses STUN for NAT traversal:
- **go2rtc** defaults to a local STUN listener (`stun:8555`), no internet required.
- **The web UI's WebRTC player** includes a fallback to Google's public STUN server (`stun:stun.l.google.com:19302`), which requires internet.
- **The web UI** uses the servers in `go2rtc.webrtc.ice_servers` for its WebRTC player and for the WebRTC connectivity check it runs when the Live view loads. If none are set, it uses Google's public STUN server (`stun:stun.l.google.com:19302`), which requires internet access from the browser. Set `ice_servers` to a STUN or TURN server on your network to avoid this.
## Home Assistant Supervisor
@@ -174,7 +179,7 @@ To run Frigate in an air-gapped or offline environment:
2.**Pre-download the training base weights**: If you plan to train custom classification models, set `TF_KERAS_MOBILENET_V2_WEIGHTS_URL` before training, then run one training job while online. Without this variable the base weights are cached outside `/config/` and are lost whenever the container is recreated, so a later training run will fail offline. If the machine never has internet access, copy the weights in manually as described below.
3.**Disable version check**: Set `telemetry.version_check: false` in your configuration.
4.**Block outbound model requests**: Set the `HF_HUB_OFFLINE=1` and `TRANSFORMERS_OFFLINE=1` environment variables to prevent HuggingFace and Transformers from attempting any network requests.
5.**Avoid cloud features**: Do not configure Frigate+, Generative AI providers that require internet, or cloud MQTT brokers.
5.**Avoid cloud features**: Do not configure Frigate+, Generative AI providers that require internet, or cloud MQTT brokers, and leave anonymous analytics off (its default).
6.**Use local model mirrors**: If limited internet is available, set the `HF_ENDPOINT`, `GITHUB_ENDPOINT`, `GITHUB_RAW_ENDPOINT`, and `TF_KERAS_MOBILENET_V2_WEIGHTS_URL` environment variables to point to local mirrors.
After these steps, Frigate will operate with no outbound internet connections.
@@ -41,7 +41,7 @@ Rockchip models are automatically converted as of 0.17. For 0.16, YOLOv9 onnx mo
## Supported detector types
Currently, Frigate+ models support CPU (`cpu`), Google Coral (`edgetpu`), OpenVino (`openvino`), ONNX (`onnx`), Hailo (`hailo8l`), and Rockchip (`rknn`) detectors.
Currently, Frigate+ models support CPU (`cpu`), Google Coral (`edgetpu`), OpenVino (`openvino`), ONNX (`onnx`), Hailo (`hailo`), and Rockchip (`rknn`) detectors.
| Hardware | Recommended Detector Type | Recommended Model Type |
@@ -21,6 +21,17 @@ class GenAIRoleEnum(str, Enum):
chat="chat"
descriptions="descriptions"
embeddings="embeddings"
transcribe="transcribe"
# Providers that can accept audio input for the transcribe role. Ollama has no
# audio input support, so claiming the role there would fail at request time.
TRANSCRIBE_CAPABLE_PROVIDERS={
GenAIProviderEnum.openai,
GenAIProviderEnum.azure_openai,
GenAIProviderEnum.gemini,
GenAIProviderEnum.llamacpp,
}
classGenAIConfig(FrigateBaseModel):
@@ -52,7 +63,7 @@ class GenAIConfig(FrigateBaseModel):
GenAIRoleEnum.chat,
],
title="Roles",
description="GenAI roles (chat, descriptions, embeddings); one provider per role.",
description="GenAI roles (chat, descriptions, embeddings, transcribe); one provider per role. Only chat, descriptions, and embeddings are granted by default; transcribe must be listed explicitly.",
)
provider_options:dict[str,Any]=Field(
default={},
@@ -66,3 +77,17 @@ class GenAIConfig(FrigateBaseModel):
description="Runtime options passed to the provider for each inference call.",
json_schema_extra={"additionalProperties":{}},
)
@model_validator(mode="after")
defvalidate_transcribe_provider(self)->Self:
"""Reject the transcribe role on providers that cannot accept audio input."""
if(
GenAIRoleEnum.transcribeinself.roles
andself.providernotinTRANSCRIBE_CAPABLE_PROVIDERS
):
raiseValueError(
f"GenAI provider '{self.provider.value}' does not support audio input "
@@ -20,6 +21,13 @@ class ImageSourceEnum(str, Enum):
recordings="recordings"
classReviewFrameModeEnum(str,Enum):
"""How review frames are presented to the GenAI provider."""
frames="frames"
annotated_frames="annotated_frames"
classReviewResponseStyleEnum(str,Enum):
"""Writing style presets for GenAI review descriptions."""
@@ -153,6 +161,11 @@ class GenAIReviewConfig(FrigateBaseModel):
description="Preferred language to request from the GenAI provider for generated responses.",
default=None,
)
frame_mode:ReviewFrameModeEnum=Field(
default=ReviewFrameModeEnum.frames,
title="Frame mode",
description="How frames are presented to the model. 'frames' sends the prompt followed by the frames, which suits models that track a sequence well on their own. 'annotated_frames' labels each frame and interleaves notes derived from object tracking, which helps models that lose track of activity that repeats or reverses.",
@@ -5,6 +5,7 @@ from pydantic import ConfigDict, Field, field_validator
from.baseimportFrigateBaseModel
__all__=[
"AudioTranscriptionModelEnum",
"CameraFaceRecognitionConfig",
"CameraLicensePlateRecognitionConfig",
"CameraAudioTranscriptionConfig",
@@ -20,6 +21,10 @@ class SemanticSearchModelEnum(str, Enum):
jinav2="jinav2"
classAudioTranscriptionModelEnum(str,Enum):
whisper="whisper"
classEnrichmentsDeviceEnum(str,Enum):
GPU="GPU"
CPU="CPU"
@@ -53,10 +58,35 @@ class AudioTranscriptionConfig(FrigateBaseModel):
description="Enable or disable automatic audio transcription for all cameras; can be overridden per-camera.",
)
language:str=Field(
default="en",
default="auto",
title="Transcription language",
description="Language code used for transcription/translation (for example 'en' for English). See https://whisper-api.com/docs/languages/ for supported language codes.",
description="Language code used for transcription/translation (for example 'en' for English), or 'auto' to let the model detect it. See https://whisper-api.com/docs/languages/ for supported language codes.",
)
model:AudioTranscriptionModelEnum|str|None=Field(
default=AudioTranscriptionModelEnum.whisper,
title="Audio transcription model or GenAI provider name",
description="The transcription backend: 'whisper' for Frigate's built-in local models, or the name of a GenAI provider with the transcribe role.",
)
@field_validator("model",mode="before")
@classmethod
defcoerce_model_enum(cls,v):
# An absent value ("model:" with nothing after it, or an explicit null)
# means unspecified, so fall back to the built-in backend. Left as None
# it would pass the GenAI-provider validation, which only inspects
# strings, and then be treated as a provider name that resolves to no
# client, turning transcription into a silent no-op.
# Explains the per-frame labels and tracker notes used by the annotated frame
# mode. Neither the notes nor this guidance say whether repeated detections are
# the same subject, since the tracking data cannot tell.
FRAME_ANNOTATION_GUIDANCE="""- Each image below is immediately preceded by a text label giving its frame number and how many seconds into the sequence it was captured. Use these labels to track the order of events and the time between them.
-Someimagesbelowareprecededbynotesfromthecamera's object tracker recording what changed at that point: an object being first detected, starting to move, reversing direction, stopping, or no longer being detected. These notes come from tracking data rather than from the images, and they are reliable. Use them to establish how many distinct activities occur and in what order, and describe every one of them."""
# once the motion is less than 5% and the number of contours is < 4, assume its calibrated
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Josh Hawkins