Rebuilding the Frigate UI for 0.14

[blakeblackshear]

TLDR: We have completely overhauled the UI for Frigate to focus on specific common user goals. It’s a big change and we hope you love it. This is just where we started, and the new UI isn’t entirely at feature parity with the current release yet.

We plan to release the first beta soon.

Background

Frigate started as a stateless object detection pipeline for live video fees with information relayed over MQTT to be used in Home Assistant. Initially, I had no intention of building a UI for it. HomeAssistant was the UI. After a few years of releases in January 2021, a community contribution with a basic web UI was made as a part of version 0.8.0. Since then, the UI has always been treated as a basic interface to expose some of the backend information that Frigate was using to detect and track objects. As new features like clips and recordings were added the UI was updated to provide an interface to view them.

A new approach

As Frigate has become more mature, it has evolved from a real time object detection pipeline into an NVR. While it still lacks some of the customization and features from NVR solutions that have existed for longer, many users are migrating from those platforms and using Frigate as their primary NVR solution. However, Frigate’s UI was never designed around specific user goals, nor was any real thought put into the efficiency of accomplishing common tasks.

Frigate+ was created in part to help fund the future development of Frigate. While still in the early phases, the initial revenue from Frigate+ subscriptions created an opportunity to hire outside resources for revamping Frigate’s UI. After looking at various options, I decided to hire Jim Designs.

The Goals

I wanted to design the new UI around specific user goals with the intent to have a UI that allows accomplishing those goals as quickly and efficiently as possible. The following goals are where we started. I expect we will continue to find better ways of accomplishing these goals in the future as well as incorporate additional goals.

What is happening right now and/or what just happened?

The first use case centers around the need to quickly see what is happening or what just happened. Imagine you hear a sound outside, your dog barks, or the doorbell rings and you want to quickly be able to see your cameras.

The new home page for Frigate is intended to be a real time dashboard for your cameras. Recent alerts are represented by animated thumbnails in a filmstrip view above your camera feeds. Your camera images update once per minute when no detectable activity is occurring to conserve bandwidth and resources. As soon as any motion is detected, cameras seamlessly switch to a live stream. In addition, camera groups have been added so you can quickly switch between indoor and outdoor cameras.

Selecting a camera takes you to a new live view where the scroll wheel can be used to zoom in and out.

What happened last night (or the past 24 hours)?

A common use case for security cameras is reviewing last night’s footage for any concerning activity. The first step of that process is checking to see what Frigate detected. In the current UI, this is accomplished by scrolling through your events. Because events are simply low level tracked objects, their time ranges may overlap or still be in progress. This makes it difficult to review a timeline of video without jumping around or watching the same section of video multiple times. If a person walks by with their dog and a car drives by at the same time, you will end up with 3 separate events all with overlapping time ranges. It is difficult to reason about what periods of time you have and haven’t reviewed.

In order to switch to a more timeline oriented review process, we implemented a new area of the app for reviewing video footage. When reviewing, Frigate intelligently identifies parts of your timeline for review by leveraging the underlying data about motion and tracked objects. These time periods are mutually exclusive, so once you have reviewed a portion of the timeline for a camera, it is marked as reviewed.

The other primary focus was on efficiency. We wanted users to be able to review these segments as efficiently as possible. Rather than scrolling through a timeline and trying to align a single point with an indication of activity, we implemented a scrollable grid of thumbnails alongside a timeline. Hovering (or swiping on mobile) over the thumbnails will start a preview playback for the segment inline. You can also grab the progress bar to scrub through the segment.

After watching a segment, it is marked as reviewed.

Lastly, not all segments are created equal. Video of people who enter your property are a different priority than those walking by on the sidewalk. By default, all person and car objects are considered alerts. You can refine what ends up in the alert tab by configuring required zones for your events.

Was anything missed last night?

In the current release, there isn’t a way to see motion activity on a timeline and watch video at that time. The ability to scrub video for things Frigate missed has been one of the most common feature requests.

Within the Motion tab of review, you can now scrub through the timeline for all cameras simultaneously with a timeline showing relative motion activity.

In order to avoid scrubbing over time periods already included in alerts and detections, you can scrub in motion only mode.

Other Notable Improvements

• Many improvements have been made to improve the usability of the mobile UI
• Live view now displays full-resolution camera streams
• Cameras with non standard aspect ratios like panoramic and portrait fit into the UI seamlessly
• Zone / mask editor has been improved to be easier to use and makes more options fully configurable in the UI
• Tune motion detection in real-time directly in the UI

[OnFreund]

This is awesome!
The designs mention that the review screen is filtered by day, which makes the "last night" uses cases a bit tricky because they technically span 2 calendar days. Is it filtered by the current date or the last 24 hours?

[NickM-27]

by default it is the last 24 hours, or a specific day can be selected on the calendar

[OnFreund]

Perfect

[blakeblackshear]

24 hours by default, then the specific day when choosing a date which would include the morning and evening of that day. It's always easy to look at last night each morning.

[nayzm]

Amazing work! Looking forward to this. 👏🏼

[nwish]

Oh man, I can't wait to jump on that beta! I'm excited about the alerts function and scrubbing!! Outstanding work guys!

Edit: When you say you plan to release the beta soon... you mean like in a few hours, right? lol

[nwish]

It is somehow possible to configure one object to alert in multiple zones, while another object alerts in one zone, and only detects in another?

[NickM-27]

yes, just set which objects each zone applies to. You can configure this in the UI by editing your zone and setting which objects apply to that zone

[wjbeckett]

Yep. I currently do this.

On my driveway camera, I have alerts set to person and car, but on my front yard camera, only person alerts where other labels are marked for detection.

The alerts and detections configuration are per camera.

[nwish]

I think I phrased my question improperly. In the UI, in individual zone settings, it seems like I can only enable or disable objects for the zone, but not for alerts and detections specifically, but I'd want to accomplish the following:

Person enters ZONE 1 and then crosses into ZONE 2 - alert for both
Car enters ZONE 1 and then crosses into ZONE 2 - detection for ZONE 1, alert for ZONE 2

So I guess I'm looking for a required_zones setting on a per object basis.

[NickM-27]

you would need 3 zones to accomplish this

zone 2, required for alerts and has person and car object
zone 1_person, required for alerts and has person object
zone_1_car, required for detections and has car object

[TorgeSchmidt]

Awesome stuff, really looking forward to it!
But what do you mean by

initial revenue from Frigate+ subscriptions

?
I setup an account, uploaded labeled images, but the option to change subscription from free to 50$/y is unavailable, marked „coming soon“…
I thought Frigate+ is not ready yet, is there a hidden button or something? 😜

[NickM-27]

Frigate+ has been available for about 8 months now, you just need to email [email protected] to request access

[binarybucks]

looks very promising. Will there be a version optimized for mobile devices?

[NickM-27]

the UI has been built with both desktop and mobile views in mind

[binarybucks]

I just tried the dev build and the new interface really is a vast improvement on both the desktop and mobile.
I really appreciate that some of the Frigate+ earnings are invested into this.

The switching from static to live images on motion is a very neat trick and superbly implemented. The loading times are also so fast that they might replace my existing professional NVR altogether and the new Motion Tuner and Frigate+ Score Range filter are two really helpful details.

Overtall, very well done and a big improvement in the right direction.

The only issue I found so far is that the interface elements on the mobile version are quite small and hard to hit at the moment .

[dz4news]

Looks very impressive and looking forward to the beta. It looks like your revenue from Frigate+ was well spent and I will gladly support you with this effort... I am one of those that completely dumped my proprietary NVR system and adopted Frigate.
Quick question... as you move forward with a standalone UI and as it matures, will the integration with HomeAssistant continue?

[NickM-27]

I don't speak for the project but personally I can see the integration sensors and services progressing but I think the media browser gets replaced with a link to the frigate ui in the sidebar

[hawkeye217]

Like @NickM-27, I am only a contributor to the project so I can't speak to future goals and direction. But the new UI does not provide automation features, notifications, or sensors. Home Assistant and Frigate work very well together, and there has been no discussion to change that whatsoever.

[dz4news]

Completely understandable.. I personally find myself using the Frigate Proxy more often to review events, etc and just use HA for the alerts and quick viewing of the cameras.

[OnFreund]

FWIW my vote would be against dropping the media browser - it's really useful and is an easy way to integrate with media players in HA (e.g. cast device)

[nomonkeynodeal]

When I saw the title I thought "oh they are finally making it possible for non-experts to configure Frigate. Maybe I will never have to type 'nano config.yml' again because I entered one wrong character and now Frigate won't start".

[NickM-27]

frigate has supported a config editor for over a year now, this config editor checks your file validity before restarting among other benefits like lint checking and auto complete. so there really is no reason to be using nano

[NickM-27]

also, the new UI does make more things configurable directly from the UI

[hawkeye217]

Motion/object masks, motion parameters, and new camera groups can be configured from the new UI without editing the config file, but configuring every parameter from frigate exclusively from the UI is a massive development undertaking. I suspect frigate will get there eventually.

[nomonkeynodeal]

And yet somehow a user with a significantly developed homelab and over a year of Frigate experience still makes configuration errors and has trouble understanding how Frigate works. What do you attribute that to exactly? Low IQ? Moral failing?

[nwish]

What do you attribute that to exactly? Low IQ? Moral failing?

Maybe just frustration getting the best of you. We've all been there. Always be ready to poke at the logs when things go south. It also took me a while to get used to Frigate's configuration.

[nomonkeynodeal]

What do you attribute that to exactly? Low IQ? Moral failing?

Maybe just frustration getting the best of you. We've all been there. Always be ready to poke at the logs when things go south. It also took me a while to get used to Frigate's configuration.

No, the answer is obviously that Frigate is difficult to configure properly and not well documented. Let me be clear that this is NOT a complaint. I'm just stating my opinion that a camera viewing GUI is low-priority compared to a configuration UI. That's it.

[shinyzard123]

I have to disagree, the inability to review motion footage is a big pain point for us and I have found Frigate's documentation to be better than many other projects that I have used. It does have a learning curve but after some time spent it is pretty straightforward in my opinion

[nwish]

A configuration UI would be nice, but obviously that wasn't part of the original set of goals, and has been set back in favor of high quality and accurate object detection. Also, did you know that you can copy/paste the example "full configuration" from the docs into your own file and start removing things that you aren't changing? If you're still having trouble, maybe start over by doing that.

[adamburgoyne]

It would be nice to have an "Add camera" wizard and camera config editor in the UI to be fair but now my cameras are setup I rarely need to touch the config whereas I do quite often review footage so I'm pretty excited for this. I'd say for most people the ability to view footage is likely to be more important.

[NickM-27]

yes, that will come in some future version

[shinyzard123]

This is awesome to see, we might fully drop Blue Iris when this launches if it solves all of our needs

[nomonkeynodeal]

A configuration UI would be nice, but obviously that wasn't part of the original set of goals, and has been set back in favor of high quality and accurate object detection. Also, did you know that you can copy/paste the example "full configuration" from the docs into your own file and start removing things that you aren't changing? If you're still having trouble, maybe start over by doing that.

I did that nearly two years ago.

[xconverge]

I have been checking in on progress/using a dev build every few weeks the past few weeks and I was always wondering/waiting for the Debug view for the camera

I just started using the latest dev build and wow, you guys have done a great job! I was blind and missing the Settings page and it is really great!

Nice work, especially @hawkeye217 and @NickM-27, I know you guys have been busy!

[nwish]

I just pulled the dev build today as well.. any pointers from your end so far? :)

[xconverge]

Nope, seems to be running pretty well! If I experience anything strange I would politely open an issue and follow best reporting practices to not be a headache for the active developers who are managing a pretty large userbase! Running a dev build I would take on additional responsibility if you can to look into problems or just revert back to stable if something is not to your liking 🤷

[nwish]

That’s my plan. I’ve got a list 😊

[NickM-27]

We do want feedback, will help to either fix things or give ideas for things that need to be emphasized in the beta release notes

[towerhand]

Great job. That looks amazing so far.

Are there any configuration changes that need to be made before switching from v0.13 to 0.14 dev?

[towerhand]

It looks like the config migration didn't like something in my config file because frigate won't start after changing to v14.

frigate  | 2024-04-28 23:36:56.823125673  *************************************************************
frigate  | 2024-04-28 23:36:56.823128853  *************************************************************
frigate  | 2024-04-28 23:36:56.823132973  ***    Your config file is not valid!                     ***
frigate  | 2024-04-28 23:36:56.823134845  ***    Please check the docs at                           ***
frigate  | 2024-04-28 23:36:56.823136042  ***    https://docs.frigate.video/configuration/index     ***
frigate  | 2024-04-28 23:36:56.823158596  *************************************************************
frigate  | 2024-04-28 23:36:56.823159823  *************************************************************
frigate  | 2024-04-28 23:36:56.823161296  ***    Config Validation Errors                           ***
frigate  | 2024-04-28 23:36:56.823162428  *************************************************************
frigate  | 2024-04-28 23:36:56.823163394  invalid literal for int() with base 10: '0.121'
frigate  | 2024-04-28 23:36:56.823889708  Traceback (most recent call last):
frigate  | 2024-04-28 23:36:56.823891717    File "/opt/frigate/frigate/app.py", line 603, in start
frigate  | 2024-04-28 23:36:56.823892859      self.init_config()
frigate  | 2024-04-28 23:36:56.823894053    File "/opt/frigate/frigate/app.py", line 136, in init_config
frigate  | 2024-04-28 23:36:56.823895194      self.config = user_config.runtime_config(self.plus_api)
frigate  | 2024-04-28 23:36:56.823896361    File "/opt/frigate/frigate/config.py", line 1456, in runtime_config
frigate  | 2024-04-28 23:36:56.823897421      camera_config.objects.filters[object] = RuntimeFilterConfig(
frigate  | 2024-04-28 23:36:56.823898522    File "/opt/frigate/frigate/config.py", line 484, in __init__
frigate  | 2024-04-28 23:36:56.823899563      mask = get_relative_coordinates(config.get("mask"), frame_shape)
frigate  | 2024-04-28 23:36:56.823900716    File "/opt/frigate/frigate/util/config.py", line 160, in get_relative_coordinates
frigate  | 2024-04-28 23:36:56.823901616      [
frigate  | 2024-04-28 23:36:56.823902717    File "/opt/frigate/frigate/util/config.py", line 161, in <listcomp>
frigate  | 2024-04-28 23:36:56.823903869      f"{round(int(points[i]) / frame_shape[1], 3)},{round(int(points[i + 1]) / frame_shape[0], 3)}"
frigate  | 2024-04-28 23:36:56.823906012  ValueError: invalid literal for int() with base 10: '0.121'
frigate  | 2024-04-28 23:36:56.823906908  
frigate  | 2024-04-28 23:36:56.823907991  *************************************************************
frigate  | 2024-04-28 23:36:56.823927703  ***    End Config Validation Errors                       ***
frigate  | 2024-04-28 23:36:56.823928905  *************************************************************

[NickM-27]

Do you have a decimal as one of your points? Otherwise I may have just missed something when working on the conversion to relative coordinates

[towerhand]

I don't see anything, but I'll send you my config via a discord DM just in case.

[NickM-27]

Perfect

[niqu1982]

This same problem. If anybody resolve this problem please describe how. My ovinf config https://pastebin.com/dmSkGepa

2024-05-11 19:02:07.549507734 invalid literal for int() with base 10: '0.0'
2024-05-11 19:02:07.549912934 Traceback (most recent call last):
2024-05-11 19:02:07.549914580 File "/opt/frigate/frigate/app.py", line 604, in start
2024-05-11 19:02:07.549915494 self.init_config()
2024-05-11 19:02:07.549916408 File "/opt/frigate/frigate/app.py", line 137, in init_config
2024-05-11 19:02:07.549917240 self.config = user_config.runtime_config(self.plus_api)
2024-05-11 19:02:07.549918129 File "/opt/frigate/frigate/config.py", line 1466, in runtime_config
2024-05-11 19:02:07.549918974 camera_config.objects.filters[object] = RuntimeFilterConfig(
2024-05-11 19:02:07.549919826 File "/opt/frigate/frigate/config.py", line 484, in init
2024-05-11 19:02:07.549920648 mask = get_relative_coordinates(config.get("mask"), frame_shape)
2024-05-11 19:02:07.549921544 File "/opt/frigate/frigate/util/config.py", line 162, in get_relative_coordinates
2024-05-11 19:02:07.549922241 [
2024-05-11 19:02:07.549923095 File "/opt/frigate/frigate/util/config.py", line 163, in
2024-05-11 19:02:07.549924326 f"{round(int(points[i]) / frame_shape[1], 3)},{round(int(points[i + 1]) / frame_shape[0], 3)}"
2024-05-11 19:02:07.549925190 ValueError: invalid literal for int() with base 10: '0.0'

[RichE85]

This looks awesome. I would like to see a password protected UI just to keep people from casually getting into the config or camera feed.

[hawkeye217]

There's already a feature request. See #1074

[wjbeckett]

With the Home Assistant/MQTT Integration, will there be different events published for an "alert" as opposed to a "detection" and "motion" events?

[NickM-27]

Naturally the messages are not 1:1

[wjbeckett]

alright, trying to play around with Notifications in HA for alerts as opposed to events.
I have the text of an alert being sent based on the alert and the object that was detected, I just can't get the thumbnail or clip of the alert. It If I try to use the after ID, it tells me the object (whether it's a thumbnail or clip) doesn't exist.

Have you implemented notifications based on the alerts part of the /reviews endpoint at all?

[NickM-27]

you can not use a review ID for the event clips / snapshots. As part of the payload there is a list of detections which is the list of relevant event IDs, those can be used. Otherwise, you can use some of the new preview.gif or preview.mp4 endpoints though this will require a slight addition to the integration

[wjbeckett]

Ooh the new preview gif and MP4 files would be nice!
I assume I need to wait for the HA Integration to be updated to utilise those?

[NickM-27]

home assistant integrations are not compiled, you only need to add a few lines to make it work blakeblackshear/frigate-hass-integration#664

[Majestic7979]

Firstly thanks so much for being constantly active here and listening to your needs. I guess I will have to eat the dog shit like I promised! lol. One thing: I absolutely love that I can scroll through the timeline and see all the events from all the cameras in sync. But what I really, really need (since I don't use events) is a timeline with a playhead where I can set the playhead at a specific place on the timeline and ALL the cameras will sync up to that specific point, then I want to press Play All and see all the cameras play in sync together. Please please please please I beg you to implement this. As I explained in a previous post, I live in Europe and it's very common to live in townhouses where your doorstep is level with the sidewalk and just have a meter of distance between your front porch and the sidewalk, with people walking by the detections are moot because an urban neighborhood of townhouses will see hundreds of people walking past the door every day just to go to work etc. So as you can imagine, in situations like that when the timeline is saturated with motion events that are merely people walking past the camera, it becomes totally irrelevant and the relevancy is more about being able to quickly play all sources in sync and without stuttering. One thing I suffer with is the stuttering when the player moves from one 10s segment to another. Every time it moves from segment A to segment B to segment C and so on, I get a half-second stutter. The more you increase the playback scale, the more stuttering. It's like a buffering effect. And if you play at 16x kiss bye bye to your sanity! As you will see is incessant stuttering as the player cannot move to the next segment seamlessly. Please for the love of almighty Jesus Christ address this major issue. It's during playback/review that any NVR shines... All NVRs can record, indeed my cameras will happily let me throw all the raw footage into a simple NFS mount. But it is the reviewing itself to find the crucial moments that makes a NVR special. I know you guys focus on detections but seriously your NVR can't be touched in terms of development speed and I sincerely hope you listen to my plight and do it 🙏🏻 God bless.

PS. I was just thinking I forgot to mention how exporting easily is equally as important as finding the relevant event in question. Currently I have to use dropdowns to select camera, whether it's live or a timelapse, and then use date/time fields. I wish I could quickly use a timeline to set in/out points just like I can do in a video editing app... put the start handle at the beginning of where I want to export and the finish handle at the end of my export, then select cameras with checkboxes, click "export" and job done, the visual timeline would help a lot. But for those who want to still be exact please include a button to "enter date and time".

[NickM-27]

PS. I was just thinking I forgot to mention how exporting easily is equally as important as finding the relevant event in question. Currently I have to use dropdowns to select camera, whether it's live or a timelapse, and then use date/time fields. I wish I could quickly use a timeline to set in/out points just like I can do in a video editing app... put the start handle at the beginning of where I want to export and the finish handle at the end of my export, then select cameras with checkboxes, click "export" and job done, the visual timeline would help a lot. But for those who want to still be exact please include a button to "enter date and time".

exporting using handles on the timeline is already implemented as well

Screen.Recording.2024-04-29.at.6.42.41.AM.mov

[NickM-27]

@Majestic7979 also one other thing to note, there is a new "loitering time" field that is added to zones. So maybe you will find that you want to use "events" (reviews) now, because you can make it an alert if a person loiters but a detection if the person just walks by. That way you get more information on the timeline of when an object was detected

[Majestic7979]

Thanks, I'm very eager to explore and see how it all works with my own footage. And I hope the development goes well :) How do I subscribe to support? Because I don't necessarily need the AI detections/models I don't subscribe to Frigate+, it would be nice to have a basic tier where I just make monthly payments to basically thank you guys for your development and customer service. It would be nice to be able to decide how much I can contribute rather than having a fixed price, that way I could vary based on an ongoing budget. While I'd like to pay for the work, some months would be less, some more, I'd rather give something rather than nothing on those months when budget is tigther.

Edit: just saw you addressing the other issues. OMG you guys are freaking amazing!!!!!!!!!!!!!!!!!!!!

[detroitgoose]

@Majestic7979 I've been super impressed with frigate+ detection. Even the base model is leaps and bounds more confident that the yolo model. A huge benefit is that you are given 12 "trainings" that allow you to tweak your model to your specific environment, and at the same time upgrade to newly released base models.

You also don't need to keep your subscription active (although it needs to be active to train new models / upgrade to new base models.) Even after you deactivate your subscription, you can still use the superior + models!

All that said, I'm really desperately waiting for improvements to the training process, but it still works ok. My issue is mostly with the difficultly in going back over images you've already trained and tweaking them. It's very clumsy at the moment, but I it'll get better.

Just giving a recommendation as someone who was very weary of "if it's worth it," and sharing personal camera images that could eventually be used in the base model. I've gotten over it. It's been pretty incredible. (the "free" models are trained on images that are far different than what a security camera sees. By crowd sourcing actual camera images, it's made a far more accurate model.)

[Majestic7979]

@Majestic7979 I've been super impressed with frigate+ detection. Even the base model is leaps and bounds more confident that the yolo model. A huge benefit is that you are given 12 "trainings" that allow you to tweak your model to your specific environment, and at the same time upgrade to newly released base models.

You also don't need to keep your subscription active (although it needs to be active to train new models / upgrade to new base models.) Even after you deactivate your subscription, you can still use the superior + models!

All that said, I'm really desperately waiting for improvements to the training process, but it still works ok. My issue is mostly with the difficultly in going back over images you've already trained and tweaking them. It's very clumsy at the moment, but I it'll get better.

Just giving a recommendation as someone who was very weary of "if it's worth it," and sharing personal camera images that could eventually be used in the base model. I've gotten over it. It's been pretty incredible. (the "free" models are trained on images that are far different than what a security camera sees. By crowd sourcing actual camera images, it's made a far more accurate model.)

Thanks for sharing your experience, I'll look into it, I don't know how it works, I don't know if I have to change my config, etc. I'll explore but it's nice to know that there's an advantage to paying apart from just saying thanks for delivering a good product and responsible support, which was my only intention :)

[Mincka]

Would it be possible to remove the fade-in / fade-out animation when footage is missing in Review?
The animation does not bring anything more than the text and it's quite annoying when you want to focus on things moving in the other cameras.

[Mincka]

That's correct. It's only in the latest hour. Otherwise it's static with "No Preview Found".

More info if relevant:
The camera has been disconnected for more than a day.

[NickM-27]

interesting... is this right at the start of the hour? do these cameras not have any motion that has happened or something? seems weird that a current hour preview would not have anything to load

[Mincka]

The cameras have been disconnected for a few days now, and new container versions were deployed in the meantime.
So they don't have any motion. Their feeds are coming from wyze-bridge when they are connected.

I click on Review, then Motion.
The 3 cameras that are disconnected are blinking continuously.
It's now 17:16 and if I go back on the timeline, it stops exactly when it goes from 16:59:00 to 16:58:30.

[NickM-27]

ah okay that makes sense

[Mincka]

Confirmed as fixed, thank you both.

[banthungprong]

Thanks for switching so fast to the new go2rtc-Version with the option to show the net of cameras with very helpful informations

[PovilasID]

I moved my main instance today and I found that there is a significant RAM increase compared to 0.13... maybe even a leak

I have an exporter that monitors frigate and here is the ram use on 0.13.2 and on the beta3 . I did a full host restart and you can see the ram use climbing back up.
Is this just a new UI overhead or is this an issue?
I also noticed a slight increase in the number of processes from 50 -> 60

[NickM-27]

I don't see any difference in RAM usage between 0.13 and 0.14. There is a known issue in 0.13 as well where in some cases the memory usage will climb due to python not freeing up memory to the system even though it has been garbage collected.

[hawkeye217]

If you're running on proxmox, there's a known issue with the kernel that manifests as a memory leak.

[PovilasID]

I am running docker container on host os

[hawkeye217]

Feel free to open a separate discussion with your config, docker compose, and screenshots of the system pages after Frigate has been running for a bit.

[PovilasID]

I will give it 24 hours or so

[Mincka]

I'm not sure of the logic behind the navigation on the alerts.

For one camera (let's call it 1), I click on the alert and I'm redirected to:
http://10.0.0.84:5000/review?group=default
I see the event with all the cameras below.

For another camera (2), if I click on an alert, I'm redirected to:
http://10.0.0.84:5000/review?group=Ext%C3%A9rieur
And I see only the cameras of this group below the event.

The issue is that cameras 1 and 2 are both in a the same group, with 2 other cameras.
I don't have other groups with these 4 cameras, so I don't understand why it's not consistent.
I also have a doubt about the behavior. I'm not sure if this always fails for camera 1 and works for camera 2, it may also have failed to redirect to the filtered view once for camera 2.

More generally, is it possible that this feature (navigation to filtered view) may lead to unexpected / unwanted behavior if a camera is present in multiple groups?

[NickM-27]

navigation when clicking an alert item from the live screen is just based off of whichever group is currently selected, not which group the camera is a part of

[Mincka]

Ok, I understand but I'm not sure to get the logic behind that.
If I'm on a camera group ("Indoor") and I click in the alert bar on a camera ("Front yard") that is not in this group, it will show cameras below that are unrelated to the event above, how is it relevant to view them with the synchronized feeds?

[kirsch33]

been wanting to test out the semantic search PRs but its merged into 0.14.1 instead of dev. i assume it wont be available from a dev build until after 14.0 is released fully then?

[NickM-27]

yes, that is correct

[Littletiger]

On the old UI, it was easy to look through events and see what zones, objects and detection confidence were detected along with the snapshot to help debug.

Is there any way to add this information to the review section currently? If not perhaps for a future update if you hover over or click the object icon in the top corner, it would display some text with that information

[NickM-27]

The review UI shows the objects that were detected and you can hover over it to see a list of all the labels

A separate view will be created to answer questions about why the review item was created

[Littletiger]

The only thing I see when hovering over it, is object names, no zones or any other information, is that correct?

[NickM-27]

Yes that's correct

[kirsch33]

is there any reason the first preview doesnt show how long ago it was like the others?

[kirsch33]

sorry for not being more clear with each of these things. @rroller is correct with what i've been trying to communicate

my humble opinion is that if #11621 is implemented it would be nice to also have them show on the top of the main page if show_reviewed or whatever varaible is true.

for me also, I have my alerts and such as very strict so I dont get very many. having them disappear after 2 hours from the main screen is quite annoying. I wish it were always a rolling quantity of the most recent 5 or so.

[NickM-27]

I am not able to reproduce the issue mentioned, all items show the time ago

[rroller]

I agree with @kirsch33

also I just experienced the missing time on the first preview. I took a screenshot. @NickM-27 can I send the screenshot to your directly?

[NickM-27]

Sure, though without some type of steps to reproduce I don't know if we'll be able to determine a fix

[kirsch33]

not sure how to provide specific steps to reproduce since its just kinda there (missing the time) when I open the app (via iOS 17 latest).

would you be interested in my config? debug logs?

[Plawasan]

Do I understand correctly the only way to export a timelapse in 0.14.0 is via an API call and not through the GUI? I'd expect a simple select in the export dialog to pick between the two options...

[NickM-27]

yes, that is correct. this is explained in the FAQ

[Plawasan]

One thing I cannot really find in the docs - can I set up multiple named timelapse_args presets for export, e.g. timelapse_25x and timelapse_100x? I don't see how to link the different arguments for export to the preset names that then show up in e.g. the options for the frigate.export_recording service in HA. I only get "realtime" and "timelapse_25x" now which seem to be the defaults.

[NickM-27]

It can only be changed in the config, multiple can not be created

[Plawasan]

Fair enough but is then there a practical reason for the _x25 in the timelapse? Would make sense to me then to only have "realtime" and "timelapse" as the options for export.

[NickM-27]

That's the default rate, when it was implemented there was no config to change it so the specifier made sense. At this point changing it would require changes to the integration and be a breaking change for something most users will never see.

[f1d094]

As I understand it, old clips/recordings are not forward compatible. What happens to them when you upgrade to 0.14?

• They are left alone and will be reaped as they reach expiration based on 'retain' settings
• They are summarily deleted from disk
• Nothing. You have to clean them up yourself.
• Something else (please elaborate)

[NickM-27]

they are left alone and deleted as they would if you kept running 0.13, nothing changes with the way the data is handled, only that the events don't show up in the UI (but recordings are still viewable)

[pyrodex]

Couple of suggestions on the Mask/Zone editor after working with some other frigate users:

1. Is it possible to MULTI-SELECT on an object mask like you can for a zone? I know I can select ALL objects or select a single one but if there is more than one I have to create an object mask for each. I know I can do it in the config by copy and pasting but would be cool if the UI did it for each selected item in the mask.
2. When using the COPY feature in the mask/zone editor how do you PASTE it into a new zone/mask?

[NickM-27]

1. this was considered but decided against I believe, because it is quite complicate. Also at least from what I have seen false positive locations often vary by label so the same mask may not work for both labels and if it did, it seems likely just choosing all labels would be better.
2. the copy is purely there for users with read-only configs where updating with the UI will not work

[Mincka]

Hello,

This is a summary of previously reported bugs / regressions / feature requests / inconsistencies still present in latest dev version.

Bugs:

1. The time picker of the export calendar is not using 24-hour format
2. Rounding fails from time to time in percentage slide on Frigate+ view - Edit: New PR with fix to confirm
3. Camera filtering panel is not updated when browsing to review with a group
4. Camera groups are not expanded when the main view is loaded using HA Proxy addon - Edit: New PR with fix to confirm

Flawed logic

1. Inconsistent navigation when clicking on an alert of a camera while seeing the cameras from a different (current) group

Regressions from 0.13

1. The lack of snapshots in the events is a regression that forces users to keep more video footage for no reason. Frigate+ view is not a fix for that.
2. Sub-label filtering was not added back to filtering, like it was done for zones

Feature requests

1. Ability to move one frame backward / forward
2. Review events with grid of camera, instead of main + thumbnails

[PovilasID]

Small inconsistency between mobile and desktop:
If you drag the timeline cursor on mobile it hides the time stamp on the cursor so you can target the motion graph more accurately but on desktop it does not, so the cursor obscures motion chart making it harder to just quickly check motion events.

Feature request:
More precise cursor: I would suggest increasing transparency of the cursor or just shrinking to a hair line while on drag for precise control or adding a "nub"/"slider" on the side permitting control. Similar to this:

Jump to motion short cut. A short cut that would jump to the next spike in motion. Spike in motion can be described by motion increase by 30-50% compared current cursor position. AI is nor a perfect and this or any other tool to quickly review the recording for "missed" detections would be grate.

[NickM-27]

Right, personally I haven't really seen that to be an issue, and I think it's good to know the exact time being looked at. We've only received this feedback a couple times so it's possible something may change in the future but I wouldn't expect it to change for 0.14.0

Once motion only is activated it hides all periods where there is no motion or there is an object detected. Then you can use the playback button or scrub the timeline and only see times of motion

[Mincka]

Playhead / cursor too big or not precise enough on desktop was indeed reported and acknowledged a few times already:
#11136 (comment)
#11136 (reply in thread)
#11136 (reply in thread)
#11136 (comment)
#11136 (comment)

The idea to put the pill at the top of the timeline on desktop when moving it is quite relevant and makes sense for consistency between mobile and desktop. Actually, it was even less an issue on mobile since you could have your thumb way on the left or right of the pill without hiding it since the timeline is taking the full width of the screen, so you could still see the time. The real issue it fixed was the fact that the pill hid small motion bars.

[OnFreund]

The playhead hiding the motion bars and review item duration is definitely still an issue.

[NickM-27]

We get feedback from all over including new discussions, Reddit, discord, email directly, etc. This is something that I only recall maybe seeing once if at all outside of the few users that have mentioned it here.

It might be helpful to understand how it's being used, when I'm scrubbing the timeline I'm looking at the cameras not the timeline because that's what has the data actually being looked for. I'm not sure why it would matter that the motion bar is covered where the scrubber is

[OnFreund]

I'm looking at the cameras, but I'm also looking at the timeline to get an overview of what's going on, things like "when will this event end?", or "when should I expect to see motion again?".
The cost-benefit here seems trivial to me - a slightly smaller or translucent playhead should not be a huge undertaking from what I understand, and doesn't involve any ongoing maintenance afterwards.

[ivanjx]

is it just me or the new web ui is extremely slow when scrolling through alerts and detections? a few dozen items is enough to cripple the scrolling. i think it is because the timeline scrollbar animation.

browsers: chromium linux, iOS safari (iphone 13)

[NickM-27]

I do see a bit of choppiness on my iPhone on a day where there are 250 review items, desktop and my Android phone are very smooth.

also why is the preview image of the item destroyed as soon as the item is out of view? whenever i tried to scroll back up all the items image being reloaded again.

not sure what you mean by "destroyed", the images are unrendered because that would be a lot of extra data / HTML to keep in memory and would slow things down even more

[ivanjx]

@NickM-27 i agree but on ios, as soon as one single item is out of the view, that very item is being unrendered. there should be some kind of buffer before deciding to unrender the out of view items.

btw this is how the performance profiler look like on vivaldi (i cant share the trace file since it contains screenshots).

[NickM-27]

I think you misunderstand, the unrendering of images is not something we are controlling, that is safari

[ivanjx]

that explains why i only see it on ios and not desktop. thanks for pointing it out.
btw here is the profiler trace without the screenshots.
Trace-20240713T193432.zip

[NickM-27]

I have tried all of my devices (MacBook, windows desktop, wife's MacBook Air) and am not seeing any desktop performance issues.

However, I worked through an issue on the safari side that we were originally working around, so the performance issues on the iPhone should be fixed with #12429

[PovilasID]

When switching from camera to camera in "history mode" sometimes the previous camera keeps playing the background.
I noticed it a couple of times because of I could hear audio from the first camera I was looking continuing on while the second one was not yet loaded.
I looked at network tab under inspect and found .ts chunks from previous getting loaded. It does not happen all the time while switching cameras but I did notice audio thing a couple times and now I confirmed it. I am sorry I can provide more specific info I find to be intermit issue.

[NickM-27]

I am not able to reproduce, but #12417 might help so that we are sure that the previous media is destroyed when a new one is loaded

[mostlychris]

Is there a way to either limit the time an alert thumbnail exists or disable that feature? I notice a spike in my CPU usage for the browser when a get a bunch of those alert thumbnails all going not to mention the repeated replay of those grabs my attention more than it should and is a bit of distraction.

[hawkeye217]

No, that's not possible right now.

But if you mark the items as reviewed from the Alerts view, they'll disappear immediately. Since they are videos, you could also right-click and select "pause" from the context menu.

[NickM-27]

if you are referring to above the live view then they only show if the alert happened within the last 2 hours. But more importantly, they only show when it has not been reviewed. If something is marked as an alert but you do not care about it enough to view it then it seems maybe it should not be configured as an alert.

In any case, you can click on the thumbnails to dismiss them and they will not show, but the point of that is to show things the user has configured as an alert

[mostlychris]

Ok makes sense. I get a ton of drive by car traffic that generates alerts. I don't want to disable alerts for vehicles because I still want to know when a vehicle enters the driveway. Zone aware thumbnails would be great. Not sure if that exists as an option or not. I know there is a way to set up a couple of zones to only alert when a vehicle crosses from one zone to a other. Not sure if that would apply to the thumbnails. I can probably also play with the loitering to limit fast moving car alerts.

[hawkeye217]

You could just create a zone for the driveway (and any other place where you want to get alerts for) and restrict Alerts to those zones. Other car traffic would end up in Detections. A loitering time setting would also be a good idea if you wanted to avoid any drive-by traffic for a specific zone.

[mostlychris]

I have two different zones (street and driveway). For some reason I wanted all cars but I don't remember why now. Changing the street zone so it's only detections will work. I'll play around with the settings. Fun part of Frigate is being able to fine tune things.