Replies: 51 comments 246 replies
-
|
Also, please don't open bug reports or support requests for development builds. At this time, what would be most helpful to us would be to keep all of your feedback here on this post. Thanks for helping test the new features! |
Beta Was this translation helpful? Give feedback.
-
|
First attempt to load latest dev build ended up with Config Editor (Safe Mode) due to this line: Commented out |
Beta Was this translation helpful? Give feedback.
-
|
Note that this seems to be the |
Beta Was this translation helpful? Give feedback.
-
|
yes, this has been fixed in the latest |
Beta Was this translation helpful? Give feedback.
-
|
Hm, I'm still seeing this issue on #85f7138 (ghcr.io/blakeblackshear/frigate:85f7138-rocm / 0.17.0-85f7138), which is currently the latest commit. |
Beta Was this translation helpful? Give feedback.
-
|
to be clear the config will be considered invalid. The issue that is fixed is that if the migration runs, it will not save that value there |
Beta Was this translation helpful? Give feedback.
-
|
Got it, thanks! |
Beta Was this translation helpful? Give feedback.
-
|
Made a config change, hit save and restart, and got this series of errors I've not noted before. Restarted 5 minutes later, got a similar error on a different camera: |
Beta Was this translation helpful? Give feedback.
-
|
that is generally fine, just issues while shutting down which did not stop Frigate from stopping |
Beta Was this translation helpful? Give feedback.
-
|
i noticed when i try to create state classification i got this logs |
Beta Was this translation helpful? Give feedback.
-
|
might be an issue with the Nvidia GPU, it works fine for me, but again I have been looking for a reason to remove it so will just do that |
Beta Was this translation helpful? Give feedback.
-
|
sure ill wait for your updated version. everyday ill always wait for you guys update before ill go to sleep wait wait to try beta and stable version we are so close to beta its looks like |
Beta Was this translation helpful? Give feedback.
-
I would expect a few more posts like this, it definitely will be a little while |
Beta Was this translation helpful? Give feedback.
-
|
i just tried new version 338b681, its says "Don't use GPU for training" but i got error log like ModuleNotFoundError: No module named 'tensorflow' |
Beta Was this translation helpful? Give feedback.
-
|
interesting, I may have missed something |
Beta Was this translation helpful? Give feedback.
-
|
On "Create New Classification", when creating a "State" type classification, it's slightly confusing to me that the different states I add are labelled "Classes". This might be technically correct but is it clearer to just says "States"? The tooltip even says "Define the different states..." 
|
Beta Was this translation helpful? Give feedback.
-
|
yeah, it is generic because classes works for objects and states, probably fine to make that dynamic |
Beta Was this translation helpful? Give feedback.
-
|
looking for better homekit guide 
if i keep frigate network host mode my frigate will keep on restart. 2nd: some problem with audio transcribe with Telugu not tried with English yet because no one speak here 0% when i press transcribe ill get logs like 2025-10-31 19:18:11 but after waiting 10min nothing happing i see my GPU and CPU will go up but then few with later no logs and nothing just silent if i press again it start no log file failed or completed. or one time ill get like only ones its not even Telugu something else 
and trying lang is te |
Beta Was this translation helpful? Give feedback.
-
|
Thanks, however we are not looking for feedback on these things currently as it says in the OP |
Beta Was this translation helpful? Give feedback.
-
|
How do you delete classification models? I'm creating quite a few tests; looks like there are a few different locations I need to delete things in (and possibly the DB?) |
Beta Was this translation helpful? Give feedback.
-
|
Yeah we haven't added that yet. Nothing in the DB, just delete the files in /clips |
Beta Was this translation helpful? Give feedback.
-
|
I am running it since a few weeks and it is rock stable.
|
Beta Was this translation helpful? Give feedback.
-
|
In that case nothing in the logs regarding the down camera: It happens for go2rtc camera and direct camera: Some process are still there for that down camera: Up camera has: |
Beta Was this translation helpful? Give feedback.
-
|
@NickM-27 Here is some early NPU feedback FWIW. I changed my 2 Openvino detectors from GPU to NPU. Inference is maybe 1ms higher than GPU. This is on a Arrow Lake 225H running Proxmox with Frigate in a LXC container. 32GB Ram, 4vcpu's. My build is a day old (brand new setup). I have 4 cameras passing 8fps of 1920*1080 h265 CBR 1.5Mb/s Ill be playing around with this build for a while yet so happy to test/break anything you want to know..... Build 0.17.0-1e50d83 ov_1: model: Vainfo: |
Beta Was this translation helpful? Give feedback.
-
|
I seem to be getting face recoginition crashes. Only thing in the logs is. |
Beta Was this translation helpful? Give feedback.
-
|
Does the output of |
Beta Was this translation helpful? Give feedback.
-
|
Yep! |
Beta Was this translation helpful? Give feedback.
-
|
Yep, your OOM killer is killing the process. Are you running in an LXC? You probably should increase the available memory. |
Beta Was this translation helpful? Give feedback.
-
|
I am using an LXC, i'll try increasing the memory |
Beta Was this translation helpful? Give feedback.
-
|
This probably goes without saying as it's been said many times, but Frigate does run best on bare metal in Docker. We see similar issues with Proxmox users all the time. |
Beta Was this translation helpful? Give feedback.
-
|
I have a 185H cpu where I got NPU to work on 0.16 with some packages inside the frigate container. Now on 0.17 it doesnt say "NPU" in the system metrics. Im using frigate+ with yolo9 detectors: 
|
Beta Was this translation helpful? Give feedback.
-
|
|
Beta Was this translation helpful? Give feedback.
-
|
Thanks, should be fixed in the next build |
Beta Was this translation helpful? Give feedback.
-
No more errors, but it doesn't seem to ever start up. seems interesting? |
Beta Was this translation helpful? Give feedback.
-
|
NPU Intereference speed example with image: ghcr.io/blakeblackshear/frigate:256817d: Setup:
Detector interference speed:
Detector CPU usage: same at both: around 1,9% The power consumption of this setup is with NPU the same like with iGPU. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks! |
Beta Was this translation helpful? Give feedback.
-
|
I've started playing with classification (state) for a iron gate which is in plain sight of one of my camera. The "creation" UI is fine but can't seem to be accessed after initial setup (should be available as Edit in "..." to change Name or add additional states. Where is the actual "state" of the particular object supposed to be found ? |
Beta Was this translation helpful? Give feedback.
-
Yes, understood, I had a look at the code in the meantime : I've added an ugly try/except clause locally and it basically seems to work (and prevent crash of the whole embeddings thread). |
Beta Was this translation helpful? Give feedback.
-
|
I've also got the same kind of error when deleting previously classified clips, as after delete the UI is not reloaded and you're still able to select the same thumbnail. |
Beta Was this translation helpful? Give feedback.
-
|
Can you elaborate, I do not see any way in the UI to delete without the UI refreshing |
Beta Was this translation helpful? Give feedback.
-
|
If I open my object page, go to one of the object classified state pages, select one or several file and hit "Delete" I get a green pop-up but the page is not reloaded and the thumbnail are still there. Selecting them again and hitting "Delete" a 2nd time throw a red pop-up and the related error in the log. |
Beta Was this translation helpful? Give feedback.
-
I've slightly modified your fix locally to the following as "files" (at least thumbnails) where actually going over "100" which is the harcoded limit : |
Beta Was this translation helpful? Give feedback.
-
|
Having a search box to filter logs would be very helpful. As would having, on mobile, tapping the top of the screen scroll the log window frame to top instead of the page itself. Right now I don't see a way to jump to top of logs on mobile like there is on desktop Also what are people most commonly doing for genai ? I'm using the recordings setting for review summaries which I know mentions rate limits but I'm hitting Google's extremely quick. I really don't want to pay for a service to use this feature but at the same time I'm not sure I have a good enough system to run ollama. I'll need to look into it |
Beta Was this translation helpful? Give feedback.
-
|
I was using a 3050 via USB4 eGPU to a mini pc and that worked well, I have since upgraded to a 5060Ti which had problems with eGPU so I built a open bench to run it. Works very well for what it is, and doesn't use much power at all when idle |
Beta Was this translation helpful? Give feedback.
-
|
I just have a 1650 super so I think I'm probably SOL to do genai locally. I'll need to look more into Googles API and see if I can get more usage somehow and/or goto the recommended review summary settings which will cut down some |
Beta Was this translation helpful? Give feedback.
-
|
You might want to look into Ollama cloud |
Beta Was this translation helpful? Give feedback.
-
|
There is definately something not right with face after build 88a2f6c. as soon as I move on from that commit faces are comming up as unknown. if i try and reprocess them i get the following. If i roll back to 88a2f6c they reprocess. 
|
Beta Was this translation helpful? Give feedback.
-
|
We've found the issue, looking into a fix |
Beta Was this translation helpful? Give feedback.
-
|
Should be fixed in the latest build. |
Beta Was this translation helpful? Give feedback.
-
|
Lastest build is working, thanks. |
Beta Was this translation helpful? Give feedback.
-
|
There is something weird still going on with the reprocess function. if you add a face and then try to reprocess the image next to it it says no faces were found until there is a new image added to the face page then i reprocesses ok. |
Beta Was this translation helpful? Give feedback.
-
|
The reason is because now the reprocess function returns correctly. Training the face model takes time so if you try to reprocess when it isn't up to date it will give that error. I will update the error to be more clear |
Beta Was this translation helpful? Give feedback.
-
|
So as a quick update on my usage of classification models: its working fairly well. I'm randomly getting errors when trying to train a new model that its failing to find images, but if I hit try again, it works. Its done in multiple times now. One other issue I'm running into now is one of the states that I thought would be nice to have (position of a manual gate) required repositioning the camera (you could see it before, but it was really tough, even for a human, and I couldn't get a classification model trained on it, I moved the camera a slightly to make it much easier), but now when I'm trying to train the model, its obviously pulling tons of images from before the repositioning. Is there a recommended way to handle camera repositioning? |
Beta Was this translation helpful? Give feedback.
-
very odd, I have never seen that, will have to see what might be going on
I would delete the images from the previous angle |
Beta Was this translation helpful? Give feedback.
-
|
Feedback on the new Review interface In short, love it. The Event/Detail views are super useful but the killer feature for me is the Object Path tracking. I think that's going to help a lot of users understand what's happening under the hood when it comes to object detection and debug configuration issues. |
Beta Was this translation helpful? Give feedback.
-
|
That's great. Glad it's working well for you. |
Beta Was this translation helpful? Give feedback.
-
|
I'm testing the tab with the "Classification" of objects and for some reason I can't train the model, when I click on the button, an endless download appears, and various errors appear in the logs from version to version. There is a version where learning takes place without problems (01452e4), but for some reason this functionality does not work on the last two versions. 
b"[concat @ 0x55d3557cbd80] Impossible to open '/tmp/cache/preview_frames/preview_Entry-1762930288.004772.webp'\n[in#0 @ 0x55d3557cba40] Error opening input: Invalid data found when processing input\nError opening input file /dev/stdin.\nError opening input files: Invalid data found when processing input\n"
|
Beta Was this translation helpful? Give feedback.
-
|
While viewing specific tags on the Explorer page and attempting to submit some snapshots to Frigate+, I encountered an issue: after submitting to Frigate+, if I close the Tracked Object Details page and scroll down, it enters an infinite loop, continuously requesting 
|
Beta Was this translation helpful? Give feedback.
-
|
Thanks, this will be fixed in the next build. |
Beta Was this translation helpful? Give feedback.
-
|
Is this correct for setting up genai? It doesnt seem to be running? |
Beta Was this translation helpful? Give feedback.
-
|
Have you enabled the review or object genai feature? That just defines the provider. |
Beta Was this translation helpful? Give feedback.
-
|
Nope, how do you do that? When I add enabled: true it say it's wrong. |
Beta Was this translation helpful? Give feedback.
-
|
See the dev documentation: https://deploy-preview-19787--frigate-docs.netlify.app/configuration/genai/genai_review |
Beta Was this translation helpful? Give feedback.
-
|
Possibly a daft question - if I want to update the activity_context_prompt per camera, do I just copy / paste the default text into the camera genai section and amend as I wish? |
Beta Was this translation helpful? Give feedback.
-
|
Yes |
Beta Was this translation helpful? Give feedback.
-
|
Should I train night b&w/infrared snapshots for classification objects or only colored snapshots? |
Beta Was this translation helpful? Give feedback.
-
|
Train under all conditions that you want your object classified for. |
Beta Was this translation helpful? Give feedback.
-
|
Is that the same for face? |
Beta Was this translation helpful? Give feedback.
-
|
If you need recognition in all conditions, then yes. |
Beta Was this translation helpful? Give feedback.
-
|
To be clear the face recognition docs recommend not to train IR images. This is because the face recognition models work off of face features not the image itself |
Beta Was this translation helpful? Give feedback.
-
|
Right, night/IR images often break invariance because:
|
Beta Was this translation helpful? Give feedback.
-
|
Currently, the Classification Models are working quite well, but I have a few minor questions:
|
Beta Was this translation helpful? Give feedback.
-
|
Beta Was this translation helpful? Give feedback.
-
|
Some kind of weird memory leak happened, I noticed a ton of swapping going on, so I checked htop and saw The last error in the logs showed this but nothing about the side camera. scrolling up I did see some errors with audio docker log |
Beta Was this translation helpful? Give feedback.
-
|
Just to followup I changed the config.yaml, go2rtc config - ffmpeg:rtsp://----#video=copy" to just - rtsp://------ I tried different audio aac, opus, I even tried h265 and h264 (base,high), different bitrates..etc, not sure how to debug why the MSE stream is failing. |
Beta Was this translation helpful? Give feedback.
-
|
This is likely a codec related issue. One reason why |
Beta Was this translation helpful? Give feedback.
-
hmm I'm not sure if it's related to audio, I actually did have #audio=copy too, just forgot to paste it in the post. For the direct rtsp stream that has broken MSE, I provide opus audio like this. Btw the camera lists its internal audio codec as G711, so I think that it needs conversion. |
Beta Was this translation helpful? Give feedback.
-
|
Yes, it definitely needs conversion. I would use |
Beta Was this translation helpful? Give feedback.
-
Ok I made some changes to my docker file, upping the shm size to 2gb and making the tmpfs 1g in size, with a 6 gb resource deploy limit, after some research when a process enters the D state (uninterruptible) it's basically blocking the system from killing it, so I am thinking it hit some kind of memory barrier causing it to run out of control,. When shm runs out, workers enter D (uninterruptible) state and balloons the VIRT. For now, I set the camera back to ffmpeg:// as before, and it's been stable for over 8 hours, the only other difference is now it's not raining and perhaps getting a more stable stream than before (less corruption). Hopefully it was the docker fixes, that made it more stable, as there is not much I can do about corruption :) . |
Beta Was this translation helpful? Give feedback.
-
|
I am currently testing the developer version [and] I find it super well-made. I am on version 097673b-rpi. I just have one question: I'm trying to set up genai with LM Studio, but it absolutely insists on an API key, yet I can't provide one. Is this normal? Here is my config: YAML environment_vars: genai: base_url: http://10.0.0.102:1234/v1 <-- Line commented out or removedapi_key: tutu Thank you for your work. |
Beta Was this translation helpful? Give feedback.
-
|
It doesn't seem like your |
Beta Was this translation helpful? Give feedback.
-
|
Yes, that's what I'm thinking, but I think it's not picking up my variable, even though it's correct. It works on v0.16. I have this in my config for both versions. How can I know if the developer version is correctly reading my variable? environment_vars: |
Beta Was this translation helpful? Give feedback.
-
|
You need to specify the environment variable in your Docker compose file. |
Beta Was this translation helpful? Give feedback.
-
|
I use this to launch my docker, I don't know how to add the variable directly when launching my docker. |
Beta Was this translation helpful? Give feedback.
-
|
I found it, thanks for your help. |
Beta Was this translation helpful? Give feedback.
-
|
Dev builds of Frigate never start up on my Jetson when configuring my streams, today I figured out why: Inside the frigate docker: root@600b8b2175f1:~$ /usr/lib/ffmpeg/jetson/bin/ffprobe -v quiet -print_format json -show_format -show_streams <my-rtsp-uri>
{
[.. hangs forever ..]
^C
root@600b8b2175f1:~$ /usr/lib/ffmpeg/jetson/bin/ffprobe -v quiet -print_format json -show_format -show_streams <my-rtsp-uri> -rtsp_transport tcp
{
"streams": [
{
"index": 0,
"codec_name": "h264",
"codec_long_name": "H.264 / AVC / MPEG-4 AVC / MPEG-4 part 10",
"profile": "Main",
"codec_type": "video",
"codec_tag_string": "[0][0][0][0]",
"codec_tag": "0x0000",
"width": 1280,
"height": 720,
"coded_width": 1280,
"coded_height": 720,
"closed_captions": 0,
"film_grain": 0,
"has_b_frames": 0,
"pix_fmt": "yuv420p",
"level": 31,
"color_range": "tv",
"color_space": "bt709",
"color_transfer": "bt709",
"color_primaries": "bt709",
"chroma_location": "left",
"field_order": "progressive",
"refs": 1,
"is_avc": "false",
"nal_length_size": "0",
"r_frame_rate": "349/12",
"avg_frame_rate": "25/1",
"time_base": "1/90000",
"start_pts": 3097,
"start_time": "0.034411",
"bits_per_raw_sample": "8",
"extradata_size": 37,
"disposition": {
"default": 0,
"dub": 0,
"original": 0,
"comment": 0,
"lyrics": 0,
"karaoke": 0,
"forced": 0,
"hearing_impaired": 0,
"visual_impaired": 0,
"clean_effects": 0,
"attached_pic": 0,
"timed_thumbnails": 0,
"captions": 0,
"descriptions": 0,
"metadata": 0,
"dependent": 0,
"still_image": 0
}
}
],
"format": {
"filename": "<my-rtsp-uri>",
"nb_streams": 1,
"nb_programs": 0,
"format_name": "rtsp",
"format_long_name": "RTSP input",
"start_time": "0.034411",
"probe_score": 100,
"tags": {
"title": "RTSP Session"
}
}
}For some reason |
Beta Was this translation helpful? Give feedback.
-
|
Interesting, I don't think I've ever seen that reported before. One workaround would be to manually specify a |
Beta Was this translation helpful? Give feedback.
-
|
I'll give that a go, thanks. I'll let you know if I ever figure out why this happens in my specific setup. |
Beta Was this translation helpful? Give feedback.
-
|
Also, Frigate tries to probe your camera first with |
Beta Was this translation helpful? Give feedback.
-
|
@hawkeye217 I investigated some more using a.o. wireshark and I actually think it makes sense that RTSP over UDP can't work inside a docker container with the default networking mode: FFmpeg sends a SETUP request: as you can see it chooses some random unused UDP ports to receive the video stream, the camera's RTSP server is probably sending the packets to my host's IP + ports specified in this setup packet, but these ports are not forwarded to the docker container since I haven't specified these ports in my docker compose file (and since they're random, that's not an option anyway). One option is to use host networking, but that means the unauthenticated port is also exposed to my LAN which is not ideal. I can see the same behaviour with For reference, docker reports |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the explanation. So the issue isn’t necessarily Docker or UDP RTSP (I know Hikvision and Axis have a specific range of UDP ports that you can map if you don't want to use host networking mode). The hang you're seeing happens because the 0.17 metadata code is calling
Edit: Actually, this change was made specifically to improve another feature (#20237), so I don't think we want to revert anything here. The code that probes your camera with |
Beta Was this translation helpful? Give feedback.
-
|
When I edit an existing camera and add the audio role, the camera is added twice in my config as two separate cameras. |
Beta Was this translation helpful? Give feedback.
-
|
What is your full config (before the change) and what camera are you trying to add the audio role to? |
Beta Was this translation helpful? Give feedback.
-
|
I am currently using only one camera. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks. The camera edit form will be removed and work on it postponed to 0.18. There's many more backend changes that need to be made to better support the feature. You should just manually edit your config for now. |
Beta Was this translation helpful? Give feedback.
-
|
Hi - Do we know if the Memryx MX3 accelerator should be working with these early 0.17 builds? If so, I am happy to test and report back |
Beta Was this translation helpful? Give feedback.
-
|
Yes, it should be. The code was contributed by the MemryX team directly. |
Beta Was this translation helpful? Give feedback.
-
OK thanks. I am doing some testing and configuring with fbf4388 ... also having to make sure all the drivers etc are installed correctly. I will report back when I have success... but good to know my errors are config-related vs. Frigate not being ready 
I have created and tested all the folders etc. needed for the MX3 to work with the Frigate container |
Beta Was this translation helpful? Give feedback.
-
|
You should follow the dev documentation: https://deploy-preview-19787--frigate-docs.netlify.app/configuration/object_detectors#memryx-mx3 |
Beta Was this translation helpful? Give feedback.
-
|
Do you have docker builds yet? I'd like to test out 17 on my 265k CPU and test the NPU |
Beta Was this translation helpful? Give feedback.
-
|
Docker builds are made available for every commit |
Beta Was this translation helpful? Give feedback.
-
|
I looked but couldn't find em. Have a link?
|
Beta Was this translation helpful? Give feedback.
-
|
https://github.com/blakeblackshear/frigate/pkgs/container/frigate |
Beta Was this translation helpful? Give feedback.
-
So I would grab this one for my intel CPU?: https://github.com/blakeblackshear/frigate/pkgs/container/frigate/579415029?tag=fbf4388-amd64 |
Beta Was this translation helpful? Give feedback.
-
|
or just |
Beta Was this translation helpful? Give feedback.
and others
Uh oh!
There was an error while loading. Please reload this page.
-
Frigate 0.17 includes many new features, some of which require additional feedback to ensure a smooth user experience. Frigate 0.17 is still under active development and is not recommended for use in production systems (there may be large / breaking bugs).
As we get closer to a beta cycle we are looking to get additional feedback from users in some specific areas so that we may get a better understanding of how things are working and fix any bugs.
We are currently looking for feedback on the following features, all other feedback will be ignored for the time being. If you test any of the following, please leave a note even if it just says that things are working well.
If you experience any bugs with config migration, or bugs with existing functionality then please give that feedback as well.
The docs are: https://deploy-preview-19787--frigate-docs.netlify.app/
Object Detectors
Nvidia: CUDA Graphs are now utilized to greatly reduce the involvement of the CPU when running inferences. This is done automatically with no config changes required. This should lead to lower minimum inference speeds and dramatically lower maximum inference speeds for Nvidia GPUs. This primarily affects object detection, but will help some other models such as face recognition. Do note that complex models such as YOLO-NAS and Semantic Search will not be supported / accelerated by CUDA Graphs.OpenVINO GPU: More subtle than Nvidia, but optimizations have been made to all OpenVINO inferences that should reduce CPU usage, memory contention (fewer number of memory transfers), and potentially lower memory usage as well.OpenVINO NPU: OpenVINO now supports NPU. Simply setNPUas the device on a supported platform and make sure/dev/accelis mapped into the container. Would be great to see inference speed examples, all model types that GPU supports should be supported by the NPU as well.Apple Silicon: Apple silicon is now supported. This requires setting up a plugin on the host so Frigate can communicate with the GPU/NPU. see the docsClassification Models
Frigate now has end-to-end support, including training, for classification models. These come in two types,
stateclassification (ex: gate open / close) andobjectclassification (ex: dog recognition). You can use the UI to create and start training the model.We are specifically looking to see examples of accuracy (good and bad) with images, any feedback on the setup wizard, bugs, etc.
https://deploy-preview-19787--frigate-docs.netlify.app/category/custom-classification
Review Item Descriptions
Frigate now supports using GenAI to summarize review items. Unlike object descriptions which are searchable, review item descriptions are structured metadata that include suspicious activity categorization. This allows Frigate UI and notifications to show the title / description along with indicators that something may require review based on the GenAI's understanding of the activity.
https://deploy-preview-19787--frigate-docs.netlify.app/configuration/genai/genai_review/
Authentication / Roles
main-user) to one of Frigate's supported roles (admin) see the docsBeta Was this translation helpful? Give feedback.
All reactions