0.16.0 Beta 4 #19114

blakeblackshear · 2025-07-12T10:53:15Z

blakeblackshear
Jul 12, 2025
Maintainer

Beta Documentation: https://deploy-preview-16390--frigate-docs.netlify.app/

Images

ghcr.io/blakeblackshear/frigate:0.16.0-beta4
ghcr.io/blakeblackshear/frigate:0.16.0-beta4-standard-arm64
ghcr.io/blakeblackshear/frigate:0.16.0-beta4-tensorrt
ghcr.io/blakeblackshear/frigate:0.16.0-beta4-rk
ghcr.io/blakeblackshear/frigate:0.16.0-beta4-rocm
ghcr.io/blakeblackshear/frigate:0.16.0-beta4-tensorrt-jp6

Changes since Beta 3

Bugfixes

Fix missing i18n keys
Hide disable option from live view context menu for view-only users
Fix global automatic live view switch
Improve support for ROCm running embedding model
Use default ffmpeg for export thumbnail to avoid case when webp support is missing
Fix LPR crash when invalid regex is provided
Increase tracked object video padding
Fix missing object snapshot / thumbnail in rare cases
Improve MQTT snapshots when required_zones is set for MQTT
Improve object label translations for objects with spaces in the label (ex: cell phone)
Don't fail on re-indexing embeddings when an image is corrupt
Improve ffmpeg handling for cameras sending corrupt stream data
Fix face recognition setting unknown incorrectly with min_faces config
Change min_faces to be based on number of recognized faces for a person
Handle config editor window resizes
Fix apple_compatibility setting from adding mutliple flags to ffmpeg
Fix review items that span over multiple day boundaries
Allow admin as default role and viewer as passed header for proxy auth
Fix file limit error for users running HAOS 16

Features

Updated ffmpeg for RockChip, including support for mjpeg hardware decoding
Tracked object sensor reset has been improved in the Frigate Integration
Slightly increase face recognition score reductions due to blur
Improve bird classification by tightening bounding box
Add "Other" category to storage metrics to indicate storage used outside of recordings
Update Hailo driver to 4.21.0. This will require updating to HA OS 16 for HA OS users
Remove TensorRT detector in favor of the onnx detector

Major Changes for 0.16.0

Breaking Changes

There are several breaking changes in this release, Frigate will attempt to update the configuration automatically. In some cases manual changes may be required. It is always recommended to back up your current config and database before upgrading:

Simply copy your current config file to a new location
Stop Frigate and make a copy of the frigate.db file

The TensorRT detector has been removed for Nvidia GPUs, the ONNX detector should be used instead. Nvidia recently pushed newer packages and dropped compatibility for older versions of TensorRT. Supporting native TensorRT detector directly has shown to be a large burden, making supporting newer hardware more difficult and increasing the difficulty in updating ONNX runtime. The decision has been made to remove the TensorRT detector which will allow us to focus on supporting more models and bringing more features to the Nvidia platform through ONNX runtime. Users who were running the TensorRT detector will need to update their configuration to use the ONNX detector, and download a new model. YOLO-NAS or YOLOv9 are recommended as they provide superior performance and accuracy compared to the models supported by TensorRT. Documentation
The go2rtc audio handling has been updated to accept any audio codec instead of only AAC. This makes it easier to get audio from cameras that use PCMA/U audio. However, if the recording ffmpeg preset preset-record-generic-audio-copy is currently used it likely will need to be changed to preset-record-generic-audio-aac to continue to support audio in recordings.
```
ffmpeg:
  ...
  output_args: 
    record: preset-record-generic-audio-aac
```
Object detection is now disabled by default. A config migrator will update your config globally with enabled: true for detect:. If you use Frigate with a read-only config, you may need to edit your own config to enable object detection manually.
```
detect:
  enabled: true
```
Frigate will no longer fall back to look for a ffmpeg binary in PATH. If you are extending the Frigate image or somehow injecting a custom ffmpeg binary in a place like /usr/local/bin, please read the custom ffmpeg build guide.
Home Assistant Add-on users will need to be aware of several important breaking changes. See the Important Notes for Home Assistant Add-on Users section below.
The base image for Frigate has been updated to Debian Bookworm. The Jetpack 4 (jp4) and Jetpack 5 (jp5) images have been removed due to not being compatible with the new base image. A new Jetson Jetpack 6 (jp6) build has been contributed by a community member.
The camera group icon pack has been updated. Some of your camera group icons may need to be re-selected.
The Hailo driver and firmware has been updated to 4.21.0. The standard Docker image now provides Hailo support. If you are using Home Assistant Operating System, version 16.0 is required. Other users will need to update the Hailo driver and firmware on their system using the user installation script.
The ROCm migraphx detector has been removed: Instead of separately supporting ROCm via MigraphX and via ONNX, the ROCm (rocm) detector has been removed, and the onnx detector should be used instead. This change should only require changing the type: rocm to type: onnx in detectors
An additional parameter has been added to movement_weights for autotracking users. A config migrator will add the new value so that Frigate starts, but if this fails or you are using relative zooming with autotracking, you should remove movement_weights from your config and recalibrate.
Proxy header mapping for user roles is now required or setting the default_role to admin is required to achieve the same functionality as previous versions. This version implements a Viewer role, which is now the default for authenticated users without a role header map defined in the config. If you are using a proxy, you must include admin in your mapped role header in order for admin users to retain their elevated privileges. See the documentation.
The strftime_fmt config option is now deprecated and will be removed in a future version. The UI will now define time formats based on your language and locale.
Some API routes are now protected and require credentials or changes. With the introduction of the Viewer role and the ability to mark review items as reviewed separately per-user, changes may be needed to any services making calls to Frigate's API.
Some users may need to adjust the tls_insecure onvif config if ONVIF PTZ controls were previously working in past versions but fail to work in 0.16. The ONVIF package was upgraded for 0.16, and several users have reported that setting tls_insecure: false fixed their issues.

Important Notes for Home Assistant Add-on Users

For Home Assistant add-on users only

The Frigate config directory for add-on users was changed from the Home Assistant config directory to the add-on config directory. With this change, the Frigate config file and database are now part of the add-on backup taken by Home Assistant, no longer needing to backup them manually. Note that only backups after 0.16 will benefit of this change. If you took a backup of Frigate 0.15 when updating to 0.16, the Frigate config file and database will not be part of it. In that case, you should still backup them manually before updating to Frigate 0.16.

Automatic migration

Upon the first Frigate 0.16 initialization, all Frigate-related files will be moved from the Home Assistant config directory to the Frigate add-on config directory. This will mean these files will no longer be accessible through your Home Assistant /config directory. They will now be found in the /addon_configs directory, which can also be accessed through the Terminal or VS Code add-ons.

In the new directory, the Frigate config file will be named config.yml rather than frigate.yml. This aligns with people running Frigate through other means. If you want to move back to Frigate 0.15 and did not backup the config file, you can copy the config.yml from the /addons_config back to /config/frigate.yml before restoring the add-on. Consider copying back other files as well, like custom model, custom ffmpeg, and custom go2rtc. Note that the database from 0.16 is not compatible with Frigate 0.15 anyway and thus should not be copied back. To restore the database, you must restore it from a manual backup.

Moving the config directory between add-on variants

The /addon_configs directory contains the config directories of all add-ons. You can move files from one addon config directory to another, like from Frigate Beta to the stable one, or vice-versa. Make sure you do that before uninstalling the add-on, unless you do not select the Also permanently delete this addon's data when uninstalling it.

Breaking change

If your Frigate config file refers to any arbitrary files in the Home Assistant config dir, like echo:/config/ scripts, their references must be manually updated to from /config/ to /homeassistant/.

New Features

Frigate 0.16 introduces several major new features.

Face Recognition

Frigate utilizes face detection and recognition to apply sub-labels to known faces. Recognized faces can be included in built-in notifications and integrated into third-party notifications.

Frigate+ is not required to use Face Recognition. However, using a Frigate+ model (or a custom model that detects face natively) enhances efficiency by eliminating the need for separate face detection per person.

The accuracy of face recognition heavily depends on the quality of the images used for training. It is highly recommended to follow the training guide to achieve the best results.

Refer to the Face Recognition documentation for more.

License Plate Recognition (LPR)

Frigate detects and recognizes license plates, saving recognized plates to tracked car objects. Known license plates can also be assigned custom labels (e.g., AB12345 --> Bob's Car).

Frigate+ is not required for License Plate Recognition. However, using a Frigate+ model (or a custom model that detects license_plate natively) improves efficiency by eliminating the need for a separate detection model.

For additional features such as regex-based matching, refer to the License Plate Recognition documentation.

Viewer Role

Frigate 0.16 introduces a Viewer role that restricts access to system features, the configuration editor, and the Frigate HTTP API. Viewers have permissions to use the Live view, Review, and Explore.

Additionally, each user can mark review items as reviewed independently so that one user's action will not hide the review items from others.

The Users tab in Settings can be used to create users and assign roles, and these roles are enforced when accessing Frigate through the authenticated port. Documentation

More fine-grained permissions (like denying access to specific cameras, for example) will come in a future version.

Proxies can also now map user roles. See the Proxy configuration documentation.

Improved Support For Object Detection Hardware and Models

Hailo8

The Hailo8 detector has been refactored by the Hailo team to greatly improve object detection performance and to add support for more model types. Documentation

ONNX & OpenVINO

Support has been added for a variety of new models for ONNX and OpenVINO detectors

Frigate now supports state-of-the-art D-FINE and RF-DETR models via the ONNX and OpenVINO detectors. Documentation
Add YOLOv3, YOLOv4, and YOLOv9 model support. Documentation
Add YOLOx model support. Documentation

RockChip (RKNN)

RockChip has had the following improvements:

GPU and NPU stats are now supported in the UI (this requires mapping /sys as read-only, see the docs for more information)
RKNN Toolkit has been updated to 2.3.2
There is now built in support for converting custom ONNX models to RKNN format
Support has been added to the RKNN detector:
- Built in support has been added for YOLOv9 models Documentation
- Built in support has been added for YOLOx models Documentation

Camera Improvements

Dynamically Enable and Disable Cameras

Frigate now offers the ability to enable / disable cameras dynamically, including buttons in the UI, MQTT, and using the camera.turn_on / camera.turn_off services in the Home Assistant Integration. Disabling a camera stops all of Frigate's decoding and processing for that camera. See the documentation.

Live Stream Selection

Frigate now allows configuring multiple streams per camera, so that any number of streams for a single camera can be created and selected in the UI. See the Live view documentation.

Camera groups now include per-camera settings to further customize the live streaming behavior of that camera in that camera group.

Other noteworthy live streaming improvements:

Create on demand recordings from the single camera Live view in the UI
Audio can be enabled on Live view dashboards through a right-click context menu
Streaming statistics make it easy to view live stream mode and performance

Other UI Changes

There are many other notable improvements to the Frigate UI:

The review timeline has been rewritten to be more performant and support zooming using pinch-to-zoom on mobile and Ctrl + scroll wheel on desktop.
Plot object paths in the Object Lifecycle pane.
Add ability to create object masks from bounding boxes in the Object Lifecycle pane by right clicking.
Add edge/point snapping to the Mask/Zone editor.
Add the earliest recording date from the database to the Storage Metrics page.
Add an underline for each day that recordings are available in the History calendar pane.
Add a scroll bar to Logs pane and stream logs live as they are recorded.
Add a Frigate+ pane to Settings to show model information and allow selection of a new model from the UI.
Add the ability to draw a rectangle on the debug view to obtain an area (in pixels and percentage) and size ratio.
Add the ability to edit sub labels directly in the Tracked Object Details pane.
Ensure the mute/unmute state is persisted for the HLS player.
Add camera selection to History view.

i18n Support

Frigate now supports internationalization and localization, making it possible to change the language of the UI. The language is auto-selected based on your browser locale and can also be changed via the Languages menu in the UI.

Some languages are not fully translated. Translation contributions are welcomed and can be made using Weblate.

Backend Features and Performance Improvements

New Base Image

The base image for Frigate has been updated to Debian 12 ("Bookworm") and Python 3.11, bringing many backend performance improvements. WARNING: The popular tteck Proxmox LXC script won't automatically support Frigate 0.16. Changes will need to be made to the script. Proxmox remains an officially unsupported installation method.

Zones and Object Area

Add the ability to estimate the speed of objects passing through zones and set a speed threshold to trigger zones. Documentation
Adds the ability to specify min_area and max_area as a decimal between 0.000001 and .99 to represent a percentage of the frame. Numbers above 1 will still represent areas in pixels. For reference, 0.000001 represents ~1 pixel at 1280x720.
The object area decimal value is now displayed wherever area pixels are displayed in the UI.

Semantic Search

Semantic Search is now optional and no longer required to be enabled to use Generative AI.
Jina V2 can now be used for Semantic Search, providing multi-lingual support. Documentation
Add the ability to trigger the reindexing of tracked object embeddings directly from the UI.

Bird Classification

Add the ability to classify detected birds and add the name to the tracked object's sub label. Documentation

ONVIF and Autotracking

Implement ONVIF reconnect if a camera goes offline.
Improve responsiveness of zooming when using relative zooming with autotracking.
Refined object tracking parameters and use re-identification when autotracking the person object.

Other notable changes

Export filenames now include the start and end date/time.
Add title as mp4 metadata to exports.
Add ability to specify cooldown time and suspend native notifications per camera. Documentation
Add an API endpoint, /recordings/summary, to return true/false for each day that recordings are available.
Implement de-duplication of log entries to prevent log flooding from the same message.
Add support for SR-IOV GPU stats.

This discussion was created from the release 0.16.0 Beta 4.

bagobones · 2025-07-12T15:09:31Z

bagobones
Jul 12, 2025

Big thank-you to all the devs on this release it is really starting to polish up nicely!

0 replies

gule2004 · 2025-07-12T15:39:40Z

gule2004
Jul 12, 2025

Now that Frigate Beta 4 uses YOLOv9, is there a model for the Hailo accelerator for Frigate+?

10 replies

NickM-27 Jul 12, 2025
Collaborator Sponsor

Wait this frigate uses yolov9?

it is just another option

Im curious how it compares vs a well trained frigate+ model.

According to the research papers they perform very similarly with yolo-nas having a slight edge
https://github.com/Deci-AI/super-gradients
https://github.com/WongKinYiu/yolov9

yousaf465 Jul 13, 2025

what about resource usage?
Yolov9 vs frigate+?

NickM-27 Jul 13, 2025
Collaborator Sponsor

It depends on which model sizes you use for each and your hardware platform.

yousaf465 Jul 13, 2025

I meant same hardware with same model size.

NickM-27 Jul 13, 2025
Collaborator Sponsor

Right, it depends on the hardware, AMD GPU for example is a big improvement. For Nvidia GPU it's very similar.

neutralvibes · 2025-07-12T15:59:06Z

neutralvibes
Jul 12, 2025

Hi chaps. I seem to have a problem with beta4 starting image: ghcr.io/blakeblackshear/frigate:0.16.0-beta4-standard-arm64. It appears at first glance that NGINX is missing a configuration element, therefore bringing everything down with it.

I get the following in the logs, which I have truncated to the section I believe is pertinent:-

frigate | 2025-07-12 15:44:50.879705849 [INFO] Starting NGINX...
frigate | 2025-07-12 15:44:54.469181780 [INFO] Preparing Frigate...
frigate | 2025-07-12 15:45:01.834220958 [INFO] Starting go2rtc healthcheck service...
frigate | s6-rc: info: service nginx successfully started
frigate | s6-rc: info: service certsync: starting
frigate | s6-rc: info: service certsync successfully started
frigate | s6-rc: info: service legacy-services: starting
frigate | 2025-07-12 15:45:22.744664648 [INFO] Starting certsync...
frigate | 2025-07-12 15:45:22.755635991 [INFO] Starting Frigate...
frigate | 2025-07-12 15:45:22.769234542 nginx: [emerg] no "events" section in configuration
frigate | 2025-07-12 15:45:22.773052087 [INFO] Preparing new go2rtc config...
frigate | 2025-07-12 15:45:22.856217040 [INFO] Service NGINX exited with code 1 (by signal 0)

Thanks for your patience.

6 replies

bagobones Jul 12, 2025

Seems like I must have upset somebody at some point, not sure either when or how. If I have, it was not intended. Am I even reporting in the right place? Reverting to beta3 works fine. Would be nice if someone would take and interest and tell me if I am missing something. I simply updated the docker image and it doesn't work for me.

Not a dev, and don't have a solution for you.. Your post is only 4 hours old on a WEEKEND.. roll back and give it a few days for a response.

hawkeye217 Jul 12, 2025
Collaborator

Seems like I must have upset somebody at some point, not sure either when or how. Have I contravened some sort of etiquette? If I have, it was not intended. Am I even reporting in the right place? Reverting to beta3 works fine. Would be nice if someone would take and interest and tell me if I am missing something. I simply updated the docker image and it doesn't work for me.

No, you haven't upset anybody. Blake, Nick, and I are just normal folks with families and other life things that occupy our time, especially on the weekends. Thanks for your report. We haven't heard of anyone else with the issue thus far. Give us a few days to see if we can reproduce the problem.

neutralvibes Jul 12, 2025

Thank you for taking the time to acknowledge this. I had already reverted to beta3 before posting originally. Hope you can enjoy the rest of the weekend.

hawkeye217 Jul 12, 2025
Collaborator

If you want to help, you could go back and find the commit that broke your specific build. Beta 3's commit was f141b58 (May 30) and beta 4 was d96efdb (July 11). Just specify image: ghcr.io/blakeblackshear/frigate:<commit_hash>-standard-arm64 where <commit_hash> is f141b58, dba9206, etc. You can skip documentation or Weblate commits.

I'd start at the builds after f141b58 (beta 3) and see which one causes it to stop working for you.

https://github.com/blakeblackshear/frigate/commits/dev/?before=d574d1edaefa0f033990664a335d49481adae016+70

neutralvibes Jul 13, 2025

Thanks, I shall give this a go.

charliejgallo · 2025-07-12T18:01:55Z

charliejgallo
Jul 12, 2025

updated frigate 0.16 beta 3 to 0.16 beta 4, Homeassistant OS 15.2 to 0S 16.0 with a Hailo-8L.... everything working as expected!

0 replies

borgqueenx · 2025-07-12T19:39:30Z

borgqueenx
Jul 12, 2025

Thanks for another release! Can i simply edit the docker compose file to the new release and restart to update?
(This is my first time updating something with docker)

1 reply

NickM-27 Jul 12, 2025
Collaborator Sponsor

Restarting (docker compose restart) won't apply the update. Instead you change the image then run docker compose up -d

bagobones · 2025-07-12T19:48:12Z

bagobones
Jul 12, 2025

@NickM-27 Thank-you for all the hard work and the changes to face recognition, I have technically been following the dev builds but as of beta4 this is getting really dialed in in terms of accuracy.

Mostly a suggestion from an end user point of view but I really think there should be a blur indicator (maybe just an icon) or the negative score for blur shown (as it is progressive now) on the faces page. When considering faces for training I am always looking at the low scoring ones and at the moment there is no way to know if it is low naturally of if it was considered blurry.

This would be very helpful for trouble shooting if blur_confidence_filter: true is helping or hurting results.

10 replies

aegjoyce Jul 17, 2025

Sorry - to confirm, if min_faces is set to 3, and recognition_threshold is 0.9, then for a person to be recognised as ‘Dave’ (sublabel assigned) there needs to be 3 frames in which ‘Dave’ is matched at a confidence level of greater than 0.9?

NickM-27 Jul 17, 2025
Collaborator Sponsor

No, there needs to be 3 facts recognized as Dave (confidence > unknown_confidence) and the weighted confidence of all of Dave's images needs to be > 0.9

bagobones Jul 17, 2025

Thanks for confirming @NickM-27 I typically have a very high number of face matches now so I wasn't quite sure.

The default 1 min_faces seems to be vulnerable to one off false positives.
2 min_faces seems to be quite robust and rarely causes a failed match as there are normally ample faces. However in my house my named faces look very similar so I get false positives
3 min_faces really forces the match to be statistically significant, however it does cause a more unlabeled events, mostly short ones.

As per the comment above I would love to see some more visual feedback to help make it clear what the "weighted confidence" was for each face.. I currently use a calculator and use the average, but I know frigate actually prefers higher resolution in the weight.

Also I am still intermittently seeing the "Unknowns" that seem to be missing an Event. I THINK some of them are caused by a frigate restart while an on going event is collecting faces but I am not really sure. I can't reproduce it

Good news is at this point I am spending much less time checking faces as nearly all matches appear to be correct now.

NickM-27 Jul 17, 2025
Collaborator Sponsor

Being honest I don't think anything is going to change for 0.16. Having the calculation shown would be more of a developer / debugging & optimization information. As a user using the face recognition feature, the calculation for how a label is chosen is not really relevant to training your face recognition model - which should be based on each face individually, not trying to match some calculation.

This would mainly serve to confuse users and provide more information that is not directly helpful to the main action that improves face recognition.

bagobones Jul 17, 2025

This would mainly serve to confuse users and provide more information that is not directly helpful to the main action that improves face recognition.

Fair

I can say the current state is MUCH more reliable than the initial one, the changes during the beta process have really dialed things in.

borgqueenx · 2025-07-12T20:32:57Z

borgqueenx
Jul 12, 2025

its weird how upgrading frigate and then starting makes all my wireless cams offline except one. (i have 8 wireless cameras in total). i rebooted the router to see what would happen, now 5 are still offline. i dont understand this behavior. the logs just say "not found".
Is this some weird behavior with go2rtc? is there perhaps a more stable older version?

3 replies

bagobones Jul 12, 2025

its weird how upgrading frigate and then starting makes all my wireless cams offline except one. (i have 8 wireless cameras in total). i rebooted the router to see what would happen, now 5 are still offline. i dont understand this behavior. the logs just say "not found". Is this some weird behavior with go2rtc? is there perhaps a more stable older version?

Restart frigate and then go check what the logs say about the connections to the cameras.

Wireless cameras can be incredibly prone to lots of intermittent disruptions and packet loss.

borgqueenx Jul 12, 2025

They report in go2rtc that they are not found. They were all fine before the update and are all fine now. I just dont get why updating frigate made them not found. The cameras itself worked in rtsp. Not in go2rtc stream.

hawkeye217 Jul 12, 2025
Collaborator

Perhaps the initial surge of wifi traffic from the cameras is causing an issue on your network? There's no technical reason why this would have changed from beta 3 to beta 4 - the go2rtc version is the same.

sgoldtho · 2025-07-13T02:36:29Z

sgoldtho
Jul 13, 2025

Seeing this message and not sure where to start looking (System BIOS, OS driver Docker config etc)...

2025-07-13 11:19:08.036987642 2025-07-13 11:19:08.114866123 2025-07-13 11:19:08.114868678 2025-07-13 11:19:08.114869640 2025-07-13 11:19:08.114870201 2025-07-13 11:19:08.114870792 2025-07-13 11:19:08.114872475 2025-07-13 11:19:08.114874128 2025-07-13 11:19:08.114874830 2025-07-13 11:19:08.114887463 2025-07-13 11:19:08.114888245 2025-07-13 11:19:08.114888886 2025-07-13 11:19:08.114889477 2025-07-13 11:19:08.114907020 2025-07-13 11:19:08.114908473 2025-07-13 11:19:08.114910156 2025-07-13 11:19:08.114911478 2025-07-13 11:19:08.114929142 2025-07-13 11:19:08.114930093 2025-07-13 11:19:08.114930985 2025-07-13 11:19:08.114931656 2025-07-13 11:19:08.114945442 2025-07-13 11:19:08.114946835 2025-07-13 11:19:08.036919165 [E:onnxruntime:, sequential_executor.cc:516 ExecuteKernel] Non-zero status code returned while running QuickGelu node. Name:'/model.0/act/Mul/QuickGeluFusion/' Status Message: HIP error hipErrorInvalidDeviceFunction:invalid device function
Process detector:onnx:
Traceback (most recent call last):
File "/usr/lib/python3.11/multiprocessing/process.py", line 314, in _bootstrap
self.run()
File "/opt/frigate/frigate/util/process.py", line 41, in run_wrapper
return run(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^
File "/usr/lib/python3.11/multiprocessing/process.py", line 108, in run
self._target(*self._args, **self._kwargs)
File "/opt/frigate/frigate/object_detection/base.py", line 136, in run_detector
detections = object_detector.detect_raw(input_frame)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/frigate/frigate/object_detection/base.py", line 86, in detect_raw
return self.detect_api.detect_raw(tensor_input=tensor_input)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/frigate/frigate/detectors/plugins/onnx.py", line 81, in detect_raw
tensor_output = self.model.run(None, {model_input_name: tensor_input})
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 266, in run
return self._sess.run(output_names, input_feed, run_options)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running QuickGelu node. Name:'/model.0/act/Mul/QuickGeluFusion/' Status Message: HIP error hipErrorInvalidDeviceFunction:invalid device function

Is it known issue?

I'm running Frigate 16 Beta 4 (ghcr.io/blakeblackshear/frigate:0.16.0-beta4-rocm) in an Unraid Docker using the onnx detector:
onnx:
type: onnx

model:
model_type: yolo-generic
width: 320
height: 320
input_tensor: nchw
input_dtype: float
path: /config/model_cache/yolov9-s.onnx
labelmap_path: /labelmap/coco-80.txt

Any guidance will be gratefully received...

7 replies

NickM-27 Jul 13, 2025
Collaborator Sponsor

your iGPU may not support the necessary features to run the model

sgoldtho Jul 13, 2025

Yeah that's what I'm thinking, is there any way to test/confirm?

sgoldtho Jul 13, 2025

docker exec -it frigate /bin/bash -c '(unset HSA_OVERRIDE_GFX_VERSION && /opt/rocm/bin/rocminfo |grep gfx)'

Gives me:
Name: gfx90c

I have environment:
HSA_OVERRIDE_GFX_VERSION: "9.0.0"

And docker exec -it frigate /opt/rocm/bin/rocminfo

Gives me:
Agent 2

Name: gfx900
Uuid: GPU-XX
Marketing Name: AMD Radeon Graphics

NickM-27 Jul 13, 2025
Collaborator Sponsor

Yeah, if it doesn't work then I don't think it will. ROCm is recommended more for newer discrete GPUs

azharkov78 Jul 14, 2025

docker exec -it frigate /bin/bash -c '(unset HSA_OVERRIDE_GFX_VERSION && /opt/rocm/bin/rocminfo |grep gfx)'

Gives me: Name: gfx90c

I have environment: HSA_OVERRIDE_GFX_VERSION: "9.0.0"

And docker exec -it frigate /opt/rocm/bin/rocminfo

Gives me: Agent 2

Name: gfx900 Uuid: GPU-XX Marketing Name: AMD Radeon Graphics

  HSA_OVERRIDE_GFX_VERSION: "11.0.0" work with 780M

yousaf465 · 2025-07-13T14:38:23Z

yousaf465
Jul 13, 2025

Image embedding with Beta4 on RTX 3060 seems to be good

but issue is searching it takes too long to search for simple term `white car with black dog`

0 replies

jtut1731 · 2025-07-14T00:15:42Z

jtut1731
Jul 14, 2025

RK3588 was detecting objects fine on 0.15, although high inference times (~50ms). Dependable object detection. I was using:

detectors: # required
  rknn: # required
    type: rknn # required
    # number of NPU cores to use
    # 0 means choose automatically
    # increase for better performance if you have a multicore NPU e.g. set to 3 on rk3588
    num_cores: 3

model: # required
  path: deci-fp16-yolonas_l
  # width and height of detection frames
  width: 320
  height: 320
  # pixel format of detection frame
  # default value is rgb but yolov models usually use bgr format
  input_pixel_format: bgr # required
  # shape of detection frame
  input_tensor: nhwc
  # Optional: path to the labelmap (default: /labelmap.txt)
  labelmap_path: /config/labelmap.txt

With 0.16-b4, I removed the model section, allowing the default model to be used. I am now getting a huge number of false positives for people with really high confidence scores. But inference times are down to ~30-35ms.

I have tried tuning the motion settings, and the person object detection thresholds. On startup, I get a ton of motion, and results in a lot of person detections. Even with those settings, I get quite a few person misses. What else should I do? Is the default model good? Also, didn't there use to be a period of time during startup that the background subtraction model was being obtained and detection wouldn't start until the model was sufficient?

7 replies

NickM-27 Jul 14, 2025
Collaborator Sponsor

There may be some issue with the latest model then, you could try downloading the older model from the repo and renaming it to see if that fixes it, otherwise you may need to try another model like YOLOX or YOLOv9

CC @MarcA711

jtut1731 Jul 14, 2025

Thanks! I tried the yolov9 with the following (from the docs):

model: # required
  path: frigate-fp16-yolov9-t
  model_type: yolo-generic
  width: 320
  height: 320
  input_tensor: nhwc
  input_dtype: float
  labelmap_path: /labelmap/coco-80.txt

And immediate improvements. No false detections on startup, none within the first 5 minutes, and a true positive (with inference speeds 30-35ms)
I haven't tried the others, but yolov9 should be the default for rockchip.

FWIW, I did NOT have model defined in my config, so the defaults were being used.

MarcA711 Jul 14, 2025

I will look into this! Thanks for reporting.

jtut1731 Jul 14, 2025

What would be the easiest way to troubleshoot this issue: Car detection on one camera was flawless until switching to B4 and frigate-fp16-yolov9-t. Now, no detection at all. Are my options to either set up a recording as a camera (after converting it to the same resolution and framerate as the detect stream), or drive a car into those areas (we don't get many cars down our street)? Am I missing an easier option?

MarcA711 Jul 17, 2025

The issue should be fixed in: #19189.
If you know how to build your own frigate image, you could do so from my fork and test if it works now. Otherwise wait for the next release and please give feedback if the problem is solved.

zeroflow · 2025-07-14T08:06:56Z

zeroflow
Jul 14, 2025
Sponsor

Device: Orin NX 16GB
Variant: 0.16.0-beta4-tensorrt-jp6

Use default ffmpeg for export thumbnail to avoid case when webp support is missing

I can confirm that generating an export now includes the thumbnail.

0 replies

def1149 · 2025-07-14T15:40:44Z

def1149
Jul 14, 2025

Just updated from 0.15.1 to 0.16.b4. No problems detected.
N100, Coral, 3 cameras

1 reply

jand5181 Jul 18, 2025

Hi. Did you need to apply any changes in configuration after this update. Have N100, Coral, 6 cams and integration with HA with 0.15.2 and would like to apply latest 16 beta version. Thanks !!

bkanuka · 2025-07-14T18:22:05Z

bkanuka
Jul 14, 2025

Hey there, I'm using a Jetson AGX Xavier which does not support Jetpack 6. Can you explain what exactly was incompatible? I would be interested in backporting to Jetpack 5.

1 reply

NickM-27 Jul 14, 2025
Collaborator Sponsor

The base image was updated to Debian Bookworm

chrispatin · 2025-07-14T18:40:39Z

chrispatin
Jul 14, 2025

The best inference time for my AMD 7600 XT was 40ms. (CPU usage was high, and had to use two detectors). Now it ranges from 12-20ms with a single detector.

Great work Nick!

2 replies

NickM-27 Jul 14, 2025
Collaborator Sponsor

to be clear did you change to a different model, or just updating to 0.16 brought this improvement?

chrispatin Jul 14, 2025

The performance increase was from upgrading from 0.16b2 to 0.16b4. Nothing in the config file was changed, other than removing the 2nd detector.

detectors:
onnx_0:
type: onnx

model:
model_type: yolonas
width: 640
height: 640

Nelis77 · 2025-07-14T20:16:59Z

Nelis77
Jul 14, 2025

Updated within HA to beta 4. Now all my faces are gone. No face library. Restored (from HA backup) to beta 3. Still no faces in the library.

2 replies

NickM-27 Jul 14, 2025
Collaborator Sponsor

Frigate does not manage or change the face library images at all, they are located in /media/frigate/clips/faces, so something must have happened to the storage mount or the image files themselves

Nelis77 Jul 14, 2025

yeah thanks. I figured that out myself already. Now looking how to fix it. Sorry for bothering you. Everything else works fine (as it seems)

TheSleepySlee · 2025-07-14T20:56:38Z

TheSleepySlee
Jul 14, 2025

Hello
Been running beta3 for some time, and updated to beta4.

faced this issue:

[E:onnxruntime:Default, env.cc:234 ThreadMain] pthread_setaffinity_np failed for thread: 734, index: 0, mask: {1, 7, }, error code: 22 error msg: Invalid argument. Specify the number of threads explicitly so the affinity is not set.

Running in LXC under proxmox (for quite some time).

0 replies

perlaursendk · 2025-07-14T21:09:21Z

perlaursendk
Jul 14, 2025

Running on Rock 5B using rockchip image
I now cant see the cameras that are using go2rtc (Reolink)

If I connect using VLC to frigate (rtsp://frigate.local:8554/forhave_sub_stream), that works nicely.
So it seems like it's not able to connect to localhost?

go2rtc:
  streams:
    have2_main_stream:
      - "ffmpeg:http://10.1.2.103/flv?port=1935&app=bcs&stream=channel0_main.bcs&user=admin&password=no-way#video=copy#audio=copy#audio=opus"
    have2_sub_stream:
      - "ffmpeg:http://10.1.2.103/flv?port=1935&app=bcs&stream=channel0_ext.bcs&user=admin&password=no-way"
    forhave_main_stream:
      - "ffmpeg:http://10.1.2.50/flv?port=1935&app=bcs&stream=channel0_main.bcs&user=admin&password=no-way#video=copy#audio=copy#audio=opus"
    forhave_sub_stream:
      - "ffmpeg:http://10.1.2.50/flv?port=1935&app=bcs&stream=channel0_ext.bcs&user=admin&password=no-way"
	  
cameras:
  forhave:
    enabled: true
    ffmpeg:
      inputs:
        - path: rtsp://127.0.0.1:8554/forhave_main_stream
          input_args: preset-rtsp-restream
          roles:
            - record
        - path: rtsp://127.0.0.1:8554/forhave_sub_stream
          input_args: preset-rtsp-restream
          roles:
            - detect

2025-07-14 20:49:46.810466221  [2025-07-14 20:49:46] watchdog.forhave               INFO    : No frames received from forhave in 20 seconds. Exiting ffmpeg...
2025-07-14 20:49:46.811439825  [2025-07-14 20:49:46] watchdog.forhave               INFO    : Waiting for ffmpeg to exit gracefully...
2025-07-14 20:49:46.944156856  [2025-07-14 20:49:46] frigate.video                  ERROR   : forhave: Unable to read frames from ffmpeg process.
2025-07-14 20:49:46.945394424  [2025-07-14 20:49:46] watchdog.forhave               ERROR   : The following ffmpeg logs include the last 100 lines prior to exit.
2025-07-14 20:49:46.945858767  [2025-07-14 20:49:46] frigate.video                  ERROR   : forhave: ffmpeg process is not running. exiting capture thread...
2025-07-14 20:49:46.946686244  [2025-07-14 20:49:46] ffmpeg.forhave.detect          ERROR   : [Parsed_scale_rkrga_0 @ 0xaaaaf885cbb0] No hw context provided on input
2025-07-14 20:49:46.947235755  [2025-07-14 20:49:46] ffmpeg.forhave.detect          ERROR   : [Parsed_scale_rkrga_0 @ 0xaaaaf885cbb0] Failed to configure output pad on Parsed_scale_rkrga_0
2025-07-14 20:49:46.947748808  [2025-07-14 20:49:46] ffmpeg.forhave.detect          ERROR   : Error initializing filters!
2025-07-14 20:49:46.948166484  [2025-07-14 20:49:46] ffmpeg.forhave.detect          ERROR   : Error marking filters as finished
2025-07-14 20:49:46.948547117  [2025-07-14 20:49:46] ffmpeg.forhave.detect          ERROR   : [out#0/rawvideo @ 0xaaaaf8815e20] Nothing was written into output file, because at least one of its streams received no packets.
2025-07-14 20:49:46.948901500  [2025-07-14 20:49:46] watchdog.forhave               INFO    : Restarting ffmpeg...

2 replies

MarcA711 Jul 17, 2025

Hey,
could you provide the output of uname -a? And also of sudo dmesg right after the crash occurred?

gwmullin Jul 18, 2025

I had the same with beta4, and started a thread as well (just to cross-link them): #19169

autumnwalker · 2025-07-15T12:38:11Z

autumnwalker
Jul 15, 2025

Running into an issue with onnx - did not have this issue on beta 3.

is:issue [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running Cast node. Name:'/pre_process/pre_process.0/Cast' Status Message: CUDA error cudaErrorNoKernelImageForDevice:no kernel image is available for execution on the device

9 replies

NickM-27 Jul 15, 2025
Collaborator Sponsor

I'll see if reducing the onnxruntime version fixes this, it's likely that the next frigate version (0.17) will drop support for some of the oldest gpus currently supported

autumnwalker Jul 15, 2025

Thank you Nick. Is there anything I can do to reduce the runtime myself or do I wait for another release?

NickM-27 Jul 15, 2025
Collaborator Sponsor

You can just wait until my PR is merged, then you can use the dev build (I will let you know the image name)

NickM-27 Jul 15, 2025
Collaborator Sponsor

you can use ghcr.io/blakeblackshear/frigate:f81ac43-tensorrt

autumnwalker Jul 15, 2025

This resolved the issue for me. Thank you Nick!

phauxcamus · 2025-07-15T20:32:16Z

phauxcamus
Jul 15, 2025

I'm running into a similar issue as autumnwalker but with the rocm variant, the two errors I get on startup before it crashes:

[ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running PRelu node. Name:'/prelu/PRelu' Status Message: HIP error hipErrorInvalidDeviceFunction:invalid device function
[ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running Cast node. Name:'/pre_process/pre_process.0/Cast' Status Message: HIP error hipErrorInvalidDeviceFunction:invalid device function

beta2 was working for me, beta3 did not (same as above). Using a Frigate+ 2025.1 yolonas model on a RX 7600.

6 replies

phauxcamus Jul 15, 2025

Doesn't look like it helped, though it's a slightly different error about libonnxruntime_providers_migraphx.so missing:

frigate  | 2025-07-15 14:20:53.544933365  [2025-07-15 14:20:53] frigate.detectors.plugins.onnx INFO    : ONNX: /config/model_cache/a487ccbd06b77b855fc0fe531099e792 loaded
frigate  | 2025-07-15 14:20:53.553846592  2025-07-15 14:20:53.553804707 [E:onnxruntime:, sequential_executor.cc:516 ExecuteKernel] Non-zero status code returned while running Cast node. Name:'/pre_process/pre_process.0/Cast' Status Message: HIP error hipErrorInvalidDeviceFunction:invalid device function
frigate  | 2025-07-15 14:20:53.749607804  *************** EP Error ***************
frigate  | 2025-07-15 14:20:53.749615396  EP Error /onnxruntime/onnxruntime/core/session/provider_bridge_ort.cc:1539 onnxruntime::Provider& onnxruntime::ProviderLibrary::Get() [ONNXRuntimeError] : 1 : FAIL : Failed to load library libonnxruntime_providers_migraphx.so with error: libonnxruntime_providers_migraphx.so: cannot open shared object file: No such file or directory
frigate  | 2025-07-15 14:20:53.749633281   when using ['MIGraphXExecutionProvider', 'ROCMExecutionProvider', 'CPUExecutionProvider']
frigate  | 2025-07-15 14:20:53.749635494  Falling back to ['ROCMExecutionProvider', 'CPUExecutionProvider'] and retrying.
frigate  | 2025-07-15 14:20:53.749637443  ****************************************
frigate  | 2025-07-15 14:20:53.749723249  Process detector:amdgpu:
frigate  | 2025-07-15 14:20:53.749727246  Traceback (most recent call last):
frigate  | 2025-07-15 14:20:53.749730173    File "/usr/lib/python3.11/multiprocessing/process.py", line 314, in _bootstrap
frigate  | 2025-07-15 14:20:53.749731686      self.run()
frigate  | 2025-07-15 14:20:53.749734022    File "/opt/frigate/frigate/util/process.py", line 41, in run_wrapper
frigate  | 2025-07-15 14:20:53.749735939      return run(*args, **kwargs)
frigate  | 2025-07-15 14:20:53.749737991             ^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:20:53.749740414    File "/usr/lib/python3.11/multiprocessing/process.py", line 108, in run
frigate  | 2025-07-15 14:20:53.749742049      self._target(*self._args, **self._kwargs)
frigate  | 2025-07-15 14:20:53.749743964    File "/opt/frigate/frigate/object_detection/base.py", line 136, in run_detector
frigate  | 2025-07-15 14:20:53.749746771      detections = object_detector.detect_raw(input_frame)
frigate  | 2025-07-15 14:20:53.749748436                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:20:53.749996201    File "/opt/frigate/frigate/object_detection/base.py", line 86, in detect_raw
frigate  | 2025-07-15 14:20:53.749998155      return self.detect_api.detect_raw(tensor_input=tensor_input)
frigate  | 2025-07-15 14:20:53.750012649             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:20:53.750014583    File "/opt/frigate/frigate/detectors/plugins/onnx.py", line 81, in detect_raw
frigate  | 2025-07-15 14:20:53.750016512      tensor_output = self.model.run(None, {model_input_name: tensor_input})
frigate  | 2025-07-15 14:20:53.750018317                      ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:20:53.750020630    File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 266, in run
frigate  | 2025-07-15 14:20:53.750022372      return self._sess.run(output_names, input_feed, run_options)
frigate  | 2025-07-15 14:20:53.750037008             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:20:53.750041515  onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running Cast node. Name:'/pre_process/pre_process.0/Cast' Status Message: HIP error hipErrorInvalidDeviceFunction:invalid device function
frigate  | 2025-07-15 14:20:54.289373286  loading data from : /config/model_cache/facedet/landmarkdet.yaml
frigate  | 2025-07-15 14:21:01.020996182  2025-07-15 14:21:01.020446232 [E:onnxruntime:, sequential_executor.cc:516 ExecuteKernel] Non-zero status code returned while running PRelu node. Name:'/prelu/PRelu' Status Message: HIP error hipErrorInvalidDeviceFunction:invalid device function
frigate  | 2025-07-15 14:21:01.167987297  Exception in thread Thread-5 (build_model):
frigate  | 2025-07-15 14:21:01.167992221  Traceback (most recent call last):
frigate  | 2025-07-15 14:21:01.167995088    File "/usr/lib/python3.11/threading.py", line 1038, in _bootstrap_inner
frigate  | 2025-07-15 14:21:01.167996844      self.run()
frigate  | 2025-07-15 14:21:01.167999067    File "/usr/lib/python3.11/threading.py", line 975, in run
frigate  | 2025-07-15 14:21:01.168001157      self._target(*self._args, **self._kwargs)
frigate  | 2025-07-15 14:21:01.168003838    File "/opt/frigate/frigate/data_processing/common/face/model.py", line 301, in build_model
frigate  | 2025-07-15 14:21:01.168005926      emb = self.face_embedder([img])[0].squeeze()
frigate  | 2025-07-15 14:21:01.168007858            ^^^^^^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:21:01.168010298    File "/opt/frigate/frigate/embeddings/onnx/base_embedding.py", line 100, in __call__
frigate  | 2025-07-15 14:21:01.168012271      outputs = self.runner.run(onnx_inputs)[0]
frigate  | 2025-07-15 14:21:01.168014269                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:21:01.168016669    File "/opt/frigate/frigate/embeddings/onnx/runner.py", line 109, in run
frigate  | 2025-07-15 14:21:01.168018604      return self.ort.run(None, input)
frigate  | 2025-07-15 14:21:01.168020424             ^^^^^^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:21:01.168023154    File "/usr/local/lib/python3.11/dist-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 266, in run
frigate  | 2025-07-15 14:21:01.168025091      return self._sess.run(output_names, input_feed, run_options)
frigate  | 2025-07-15 14:21:01.168027704             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
frigate  | 2025-07-15 14:21:01.168031146  onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Non-zero status code returned while running PRelu node. Name:'/prelu/PRelu' Status Message: HIP error hipErrorInvalidDeviceFunction:invalid device function
frigate  | 2025-07-15 14:21:09.594433938  [2025-07-15 14:21:09] frigate.watchdog               INFO    : Detection appears to be stuck. Restarting detection process...

NickM-27 Jul 15, 2025
Collaborator Sponsor

Oh, I just realized the error you are showing is based on the face model. It may be the case that ROCm does not support the large face recognition model and you need to use the small model

phauxcamus Jul 15, 2025

Tried switching to the smaller face model and no change, as well as tried the 320 Yolo model (instead of 640) with no change.

NickM-27 Jul 15, 2025
Collaborator Sponsor

will need to see a full copy of logs

phauxcamus Jul 15, 2025

Sure, see attached. I can crank up verbosity if you need it

frigate.log

marcosvfc · 2025-07-16T02:19:36Z

marcosvfc
Jul 16, 2025

I notice the new On Demand Record on 0.16 feature but I'm little bit concerned about something. If a user starts a On Demand Record and forget to close/stop that recording, will it run "forever"? You know, until the storage fill up, reaches its capacity. Is there a max time for recordings?

3 replies

NickM-27 Jul 16, 2025
Collaborator Sponsor

Frigate's storage recovery will kick in just the same. I believe whenever the frontend is unloaded the manual recording is stopped anyway

hawkeye217 Jul 16, 2025
Collaborator

Correct. The frontend is set up to stop the on-demand recording if you navigate away from the page or close the browser.

marcosvfc Jul 16, 2025

OK, I get it, you are right. I'll keeping making some tests with this version. Thank you all!

dreamerpit · 2025-07-16T07:15:48Z

dreamerpit
Jul 16, 2025

Hi all
had beta 3 running and updated to beta 4 with raspi 5 and hailo8. No issues during update.
Thanks to all involved developers and supporting people again....
Only a few small things:
I found some config snippet from the project, which so far as I understand, can be added to normal config file as test cam and would allow to see a closed loop mp4 for debugging.

1.)

cameras:
  test:
    ffmpeg:
      inputs:
        - path: /media/frigate/car-stopping.mp4
          input_args: -re -stream_loop -1 -fflags +genpts
          roles:
            - detect
            - rtmp
    detect:
      height: 1080
      width: 1920
      fps: 5

Is it possible as mentioned above or am I wrong ?

2.) The beta 4 log file has some unknown entries, which have been also in beta 3.

2025-07-16 08:50:31.013875892  [INFO] Preparing Frigate...
2025-07-16 08:50:31.997654414  [INFO] Starting Frigate...
2025-07-16 08:50:33.299871841  /usr/local/lib/python3.11/dist-packages/numpy/core/getlimits.py:549: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
2025-07-16 08:50:33.299877915    setattr(self, word, getattr(machar, word).flat[0])
2025-07-16 08:50:33.300098785  /usr/local/lib/python3.11/dist-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
2025-07-16 08:50:33.300104396    return self._float_to_str(self.smallest_subnormal)
2025-07-16 08:50:33.300106248  /usr/local/lib/python3.11/dist-packages/numpy/core/getlimits.py:549: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
2025-07-16 08:50:33.300110877    setattr(self, word, getattr(machar, word).flat[0])
2025-07-16 08:50:33.300112748  /usr/local/lib/python3.11/dist-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
2025-07-16 08:50:33.300114785    return self._float_to_str(self.smallest_subnormal)
2025-07-16 08:50:34.606712604  [2025-07-16 08:50:34] frigate.util.config            INFO    : Checking if frigate config needs migration...
2025-07-16 08:50:34.646045094  [2025-07-16 08:50:34] frigate.util.config            INFO    : frigate config does not need migration...
2025-07-16 08:50:34.814537569  [2025-07-16 08:50:34] frigate.app                    INFO    : Starting Frigate (0.16.0-d96efdb)
2025-07-16 08:50:34.826231564  [2025-07-16 08:50:34] peewee_migrate.logs            INFO    : Starting migrations
2025-07-16 08:50:34.826727525  [2025-07-16 08:50:34] peewee_migrate.logs            INFO    : There is nothing to migrate
2025-07-16 08:50:34.912663701  [2025-07-16 08:50:34] frigate.app                    INFO    : Recording process started: 379
2025-07-16 08:50:34.922252797  [2025-07-16 08:50:34] frigate.app                    INFO    : Review process started: 382
2025-07-16 08:50:34.926531874  [2025-07-16 08:50:34] frigate.app                    INFO    : go2rtc process pid: 122
2025-07-16 08:50:34.943591721  [2025-07-16 08:50:34] detector.hailo                 INFO    : Starting detection process: 404
2025-07-16 08:50:34.953305890  [2025-07-16 08:50:34] frigate.app                    INFO    : Output process started: 422
2025-07-16 08:50:35.386371833  [2025-07-16 08:50:35] frigate.app                    INFO    : Camera processor started for Cam1: 467
2025-07-16 08:50:35.386950312  [2025-07-16 08:50:35] frigate.app                    INFO    : Camera processor started for Cam2: 470
2025-07-16 08:50:35.390301356  [2025-07-16 08:50:35] frigate.app                    INFO    : Camera processor not started for disabled camera Cam3
2025-07-16 08:50:35.436935042  [2025-07-16 08:50:35] frigate.app                    INFO    : Capture process started for Cam1: 496
2025-07-16 08:50:35.454417202  [2025-07-16 08:50:35] frigate.app                    INFO    : Capture process started for Cam2: 502
2025-07-16 08:50:35.477184528  [2025-07-16 08:50:35] frigate.app                    INFO    : Capture process not started for disabled camera Cam3
2025-07-16 08:50:35.575739863  [2025-07-16 08:50:35] frigate.api.fastapi_app        INFO    : Starting FastAPI app
2025-07-16 08:50:35.983234915  [2025-07-16 08:50:35] frigate.api.fastapi_app        INFO    : FastAPI started

3.)
Only an idea... Would it be possible to have the working model name visible in the system statistics ?

Many thanks

8 replies

dreamerpit Jul 18, 2025

restart: unless-stopped
stop_grace_period: 30s

privileged: true

image: ghcr.io/blakeblackshear/frigate:stable-h8l

image: ghcr.io/blakeblackshear/frigate:0.16.0-beta3-standard-arm64

image: ghcr.io/blakeblackshear/frigate:0.16.0-beta3

image: ghcr.io/blakeblackshear/frigate:0.16.0-beta4
shm_size: 2048mb
devices:
  - /dev/hailo0
  - /dev/media0:/dev/media0
  - /dev/media1:/dev/media1
  - /dev/media2:/dev/media2
  - /dev/video19:/dev/video19
volumes:
  - /etc/localtime:/etc/localtime:ro
  - ./config:/config
  - ./storage:/media/frigate
  - type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
    target: /tmp/cache
    tmpfs:
      size: 1000000000
ports:
  - "8971:8971"
  - "8554:8554" # RTSP feeds
  - "1984:1984" # Web-Interface go2rtc
  - "1935:1935" # RTMP feeds
  - "8555:8555/tcp" # WebRTC over tcp
  - "8555:8555/udp" # WebRTC over udp

MarcA711 Jul 18, 2025

So you will need to place a file called 'Cam1_20250607_214615-20250607_214926_ptc5lo.mp4' in ./storage. Since you are using relative paths, make sure that you place it in the correct location.

If it still doesn't work, check if the file is correctly mounted in your container. Use docker ps to find out your container id, and use docker exec <Container id> ls -al /media/frigate. Do you See your file?

dreamerpit Jul 18, 2025

do not see the file. That exactly my problem. In the home directory are these directorys: .config, config, storage, but not .storage. I am a bit confused in the moment. Docker is relative new for me...

MarcA711 Jul 18, 2025

Why should there be .storage? storage is the correct directory, place your file there.

dreamerpit Jul 18, 2025

that was the right way. It works now. I think its time now for me to go in the details with docker.
Thank you very much.

fahnoe · 2025-07-16T14:11:48Z

fahnoe
Jul 16, 2025

I started my Frigate journey with 0.16.0 Beta 3 on a Raspi 5 with Hailo-8 (actually a Seeed Studio reComputer AI R2130-12 which includes the 8G Pi5, Hailo-8, PCIe3.0 Switch to dual M.2, case, cooler) using older Axis cams.

Although outside the beta, one comment I'd add regards Frigate's Hailo user_installation.sh as it is looking for linux-modules-extra-$(uname -r) on all but x86_64. This probably should also include aarch64 as well.

https://github.com/blakeblackshear/frigate/blob/dev/docker/hailo8l/user_installation.sh

arch=$(uname -m)

if [[ $arch == "x86_64" ]]; then
    sudo apt install -y linux-headers-$(uname -r);
else
    sudo apt install -y linux-modules-extra-$(uname -r);
fi

root@pieye:/opt/appdata/frigate# uname -m
aarch64

root@pieye:/opt/appdata/frigate# apt list linux-headers-$(uname -r)
Listing... Done
linux-headers-6.12.34+rpt-rpi-2712/stable,now 1:6.12.34-1+rpt1~bookworm arm64 [installed,automatic]

I also notice the following UserWarnings from numpy's getlimits.py in both Beta 3 and Beta 4:

frigate  | 2025-07-16 08:26:47.758439848  [INFO] Starting Frigate...
frigate  | 2025-07-16 08:26:49.261676451  /usr/local/lib/python3.11/dist-packages/numpy/core/getlimits.py:549: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
frigate  | 2025-07-16 08:26:49.261681896    setattr(self, word, getattr(machar, word).flat[0])
frigate  | 2025-07-16 08:26:49.261684618  /usr/local/lib/python3.11/dist-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float64'> type is zero.
frigate  | 2025-07-16 08:26:49.261686544    return self._float_to_str(self.smallest_subnormal)
frigate  | 2025-07-16 08:26:49.261688488  /usr/local/lib/python3.11/dist-packages/numpy/core/getlimits.py:549: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
frigate  | 2025-07-16 08:26:49.261690581    setattr(self, word, getattr(machar, word).flat[0])
frigate  | 2025-07-16 08:26:49.261692396  /usr/local/lib/python3.11/dist-packages/numpy/core/getlimits.py:89: UserWarning: The value of the smallest subnormal for <class 'numpy.float32'> type is zero.
frigate  | 2025-07-16 08:26:49.261694285    return self._float_to_str(self.smallest_subnormal)

Other than those minor comments, two thumbs up for a remarkable collection of software! Frigate is a heck of an upgrade from the Zoneminder that I've been running for a decade+.

--Larry

2 replies

NickM-27 Jul 16, 2025
Collaborator Sponsor

I'd add regards Frigate's Hailo user_installation.sh as it is looking for linux-modules-extra-$(uname -r) on all but x86_64. This probably should also include aarch64 as well.

it depends what OS you are using, for many users this fails on aarch64, which is the reason for the alternative installation

fahnoe Jul 16, 2025

Fair enough, just passing on that I noted the error from user_installation.sh:

Reading package lists... Done
Building dependency tree... Done
Reading state information... Done
E: Unable to locate package linux-modules-extra-6.12.34+rpt-rpi-2712
E: Couldn't find any package by glob 'linux-modules-extra-6.12.34+rpt-rpi-2712'

--Larry

opentinker · 2025-07-17T09:43:08Z

opentinker
Jul 17, 2025

I've also succesfully upgraded Frigate to 0.16.0 Beta 4 on my Raspi 5 with a Hailo 8L.

Camera setup:

3x PoE Cams with main stream 3072x1728 @ 20 fps (Annke NC500)
1x PoE Cam with main stream 3840x2160 @ 20 fps (Annke NC800)
1x Wifi Cam with main stream 2560x1440 @ 20 fps (Reolink E1 Outdoor)
4x Wifi Cams with main stream 1280x720 @ low fps (ESP32-Cam)

All cameras record and stream into YOLOv6n for detection and classification. System status (CPU etc.) is green! :-)

Regarding the Hailo version mismatches: I wasted time on version hell before realizing you have to keep three components in sync.
Here’s the breakdown from my understanding:

Library (used by the actual application, e.g. Frigate or HailoRT CLI)
Driver
Firmware (AKA 'device')

Library Version:

The application library version is what matters. For Frigate, the container image defines this (currently v4.21.0). Driver and firmware must match, even if newer versions are out there.

Other tools (like HailoRT CLI) need to match as well - otherwise you get errors. Frigate itself doesn’t care if HailoRT is behind, but HailoRT won’t work:

pi@raspi-03:~ $ hailortcli fw-control identify
[HailoRT] [error] CHECK failed - Driver version (4.21.0) is different from library version (4.20.0)
[HailoRT] [error] Driver version mismatch, status HAILO_INVALID_DRIVER_VERSION(76)
[HailoRT] [error] CHECK_SUCCESS failed with status=HAILO_INVALID_DRIVER_VERSION(76)
[HailoRT] [error] CHECK_SUCCESS failed with status=HAILO_INVALID_DRIVER_VERSION(76)
[HailoRT] [error] CHECK_SUCCESS failed with status=HAILO_INVALID_DRIVER_VERSION(76)
[HailoRT CLI] [error] CHECK_SUCCESS failed with status=HAILO_INVALID_DRIVER_VERSION(76)
pi@raspi-03:~ $

Driver Version:

The PCI driver hailo_pci comes as source code from Hailo and must be compiled for your target system.

Check the installed driver version with modinfo:

pi@raspi-03:~ $ modinfo hailo_pci
filename:       /lib/modules/6.12.34+rpt-rpi-2712/kernel/drivers/misc/hailo_pci.ko.xz
version:        4.21.0
license:        GPL v2
description:    Hailo PCIe driver
author:         Hailo Technologies Ltd.
import_ns:      DMA_BUF
srcversion:     2608277ECA802FBC0F58FFD
alias:          pci:v00001E60d000026A2sv*sd*bc*sc*i*
alias:          pci:v00001E60d000043A2sv*sd*bc*sc*i*
alias:          pci:v00001E60d000045C4sv*sd*bc*sc*i*
alias:          pci:v00001E60d00002864sv*sd*bc*sc*i*
depends:        
name:           hailo_pci
vermagic:       6.12.34+rpt-rpi-2712 SMP preempt mod_unload modversions aarch64
parm:           o_dbg:int
parm:           no_power_mode:Disables automatic D0->D3 PCIe transactions (invbool)
parm:           force_allocation_from_driver:Determines whether to force buffer allocation from driver or userspace (int)
parm:           force_desc_page_size:Determines the maximum DMA descriptor page size (must be a power of 2) (int)
parm:           force_hailo10h_legacy_mode:Forces work with Hailo10h in legacy mode(relevant for emulators) (bool)
parm:           force_boot_linux_from_eemc:Boot the linux image from eemc (Requires special Image) (bool)
parm:           support_soft_reset:enables driver reload to reload a new firmware as well (bool)
pi@raspi-03:~ $

Firmware Version:

Firmware from Hailo goes in /lib/firmware/hailo/. The active firmware is set by the hailo8_fw.bin symlink:

pi@raspi-03:~ $ ll /lib/firmware/hailo
total 984
-rw-r--r-- 1 root root 164224 Jan 27 19:20 hailo8_fw.4.17.0.bin
-rw-r--r-- 1 root root 165816 Jan 27 19:20 hailo8_fw.4.18.0.bin
-rw-r--r-- 1 root root 166260 Sep 26  2024 hailo8_fw.4.19.0.bin
-rw-r--r-- 1 root root 166116 Dec 24  2024 hailo8_fw.4.20.0.bin
-rw-r--r-- 1 root root 166116 Jan 12  2025 hailo8_fw.4.20.1.bin
-rw-r--r-- 1 root root 164512 May 25 21:07 hailo8_fw.4.21.0.bin
lrwxrwxrwx 1 root root     20 Jul 13 15:32 hailo8_fw.bin -> hailo8_fw.4.21.0.bin
pi@raspi-03:~ $

Hope, this is all correct and helps for a basic understanding!

0 replies

dreamerpit · 2025-07-17T16:29:32Z

dreamerpit
Jul 17, 2025

Hi all
can anybody please comment my post from yesterday ?
thanks...

0 replies

bagobones · 2025-07-18T16:44:33Z

bagobones
Jul 18, 2025

Observation but with Face tracking 0.16 it has highlighted at least in my configuration that object tracking persistence is quite poor.

IE a person object staying in view of the camera loses tracking and starts over quite frequently.. This is very apparent when they are being labeled with peoples names as the name label is lost and has to start over with a new face detection.

If for example I go and pull up all videos with a specific name in them they will often contain huge gaps where the tracking was lost and no label was applied.

This isn't a criticism but an observation that has been further highlighted by the new features, the events still record because a person was detected, those time period just don't have a sub label (particularly if they sit down or lay down etc)

0 replies

yousaf465 · 2025-07-19T09:58:22Z

yousaf465
Jul 19, 2025

How to search for all motorcycles with a detected number plate using Frigate+? The semantic method doesn't work, even though motorcycles with license plates are detected.

3 replies

NickM-27 Jul 19, 2025
Collaborator Sponsor

You can search for all motorcycles then look for any recognized plate. There's no way to filter on just a detected plate that didn't have characters recognized

yousaf465 Jul 19, 2025

I want to list all motorcycles with recognised number plates.
Don't want to manually looking at all images with unrecognised number plates.

NickM-27 Jul 19, 2025
Collaborator Sponsor

For now you need to use the license plate filter and select the plates in the more filters sidebar, an easier way to just look for any object with a license plate recognized will come in a future version

TiddlyWiddly · 2025-07-19T16:12:02Z

TiddlyWiddly
Jul 19, 2025

Is the /plus endpoint gone or moved or something for submitting images for labeling? It worked before I upgraded to this beta. I see the new UI settings tab for plus which recognizes my API key and model but I don't see how to submit images for training (except for facial).

2 replies

NickM-27 Jul 19, 2025
Collaborator Sponsor

This was a change in 0.15, you should use the explore view. See the FAQ for 0.15 #16391 Usage Question 1

TiddlyWiddly Jul 19, 2025

I had missed that, got it now. Thanks Nick for your help.

bagobones · 2025-07-20T00:31:46Z

bagobones
Jul 20, 2025

Just because it is funny

0 replies

Uh oh!

0.16.0 Beta 4 #19114

Uh oh!

blakeblackshear Jul 12, 2025 Maintainer

Images

Changes since Beta 3

Bugfixes

Features

Major Changes for 0.16.0

Breaking Changes

Important Notes for Home Assistant Add-on Users

New Features

Face Recognition

License Plate Recognition (LPR)

Viewer Role

Improved Support For Object Detection Hardware and Models

Hailo8

ONNX & OpenVINO

RockChip (RKNN)

Camera Improvements

Dynamically Enable and Disable Cameras

Live Stream Selection

Other UI Changes

i18n Support

Backend Features and Performance Improvements

New Base Image

Zones and Object Area

Semantic Search

Bird Classification

ONVIF and Autotracking

Other notable changes

Replies: 28 comments · 85 replies

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

NickM-27 Jul 12, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

NickM-27 Jul 13, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

NickM-27 Jul 13, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

Uh oh!

Uh oh!

hawkeye217 Jul 12, 2025 Collaborator

Uh oh!

Uh oh!

hawkeye217 Jul 12, 2025 Collaborator

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

NickM-27 Jul 12, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

Uh oh!

Uh oh!

NickM-27 Jul 17, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

NickM-27 Jul 17, 2025 Collaborator Sponsor

Uh oh!

Uh oh!

blakeblackshear
Jul 12, 2025
Maintainer

Replies: 28 comments 85 replies

NickM-27 Jul 12, 2025
Collaborator Sponsor

NickM-27 Jul 13, 2025
Collaborator Sponsor

NickM-27 Jul 13, 2025
Collaborator Sponsor

hawkeye217 Jul 12, 2025
Collaborator

hawkeye217 Jul 12, 2025
Collaborator

NickM-27 Jul 12, 2025
Collaborator Sponsor

NickM-27 Jul 17, 2025
Collaborator Sponsor

NickM-27 Jul 17, 2025
Collaborator Sponsor