I love frigate... just had my neighbors come by because their dog was sick and were wondering if the dog had got into something in our back yard. Pulled up frigate, searched for "black dog" on the backyard camera, and found all the video of their dog.
So the app is free to download from the Apple site, and will run free, and is open source, but you have in-app purchases, and certain features can’t be used until you pay for them, is that right?
What are the paid features and what are the costs? Do I have to install the app to see the list of paid features and costs?
You might get a better response from HN if you give us more info up front.
Paid features are Live and event clip viewing over the internet, and receiving iOS notifications. You're paying for use of my server in those cases though, not for features I've made closed source. You can edit the code to use your own server if you wish too.
I'm new to HN and thought shilling the paid stuff violates the rules, so I didn't mention them.
(I'm a mod here) - it's fine to talk about paid features, as long as it's clear which ones are paid and which ones not.
The only thing that wouldn't be fine is to post a Show HN with no way to try the product out (https://news.ycombinator.com/showhn.html) and you're fine on that part.
Am I reading your README correct, that in order to sign up to use the app on Android, you have to install and sign up using an iOS device (using Apple's payment system) and then login on Android using the credentials you created?
Yeah sorry that’s confusing, I need to change or remove it until I’ve a payment system setup.
There is an unfinished but functional APK and android project in the repo, but it’s not on the Google Play store yet, their approval process for new individual devs is long
Anyone can recommend a good quality camera without spyware and ideally open sw stack.
I am willing to do it myself with little soldering etc.
that’s one rabbit hole didn’t enter yet
Depending on your definition of "good quality", you might find this project useful: https://thingino.com/
Most cameras on that list are low cost, typically with 4-5MP sensors. They don't compete on the high end in terms of image quality but you will have an open source firmware stack with root access over SSH.
Models from Eufy, Cinnado, Jooan, TP-Link, WUUK, Galayou are relatively easy to source on Amazon or Aliexpress.
The best option is just to assume any IPCam is unsafe and firewall them off in my experience; even with a fully open source camera stack connecting it directly to internet is not that great a practice. Put them on a no internet access VLAN and you can largely buy whatever cheap IPCams you want, etc etc. If you want remote access you should expose the server running the camera management software/NVR securely, not the cams.
This is basically how I run Frigate at home today, with only the NVR able to reach the camera IPs on my no web access “internet of nothing” VLAN.
It's not open source but used Axis cameras are pretty cheap and have rtsp and onvif support. Those mostly come from commercial installs and can be configured offline using a web interface.
Nothing good has an open software stack. There are some brands (eg: Axis, Bosch, Hanwha), that support 3rd party apps that can run on the camera and perform various tasks, including AI applications.
Any product that would fall under the good quality segment is primarily targeted at the commercial market, and nobody there is looking for open software.
There's https://openipc.org/ , if open source camera firmware is of interest to you. I actually ordered a few supported IP camera modules (basically complete IP cameras but without the case) from Aliexpress and tested that I'm able to compile a firmware, I shall see if I get it working once they arrive.
It's not quite clear to me what the firmware is actually able to do, though. Apparently its motion detection is very basic, though, so you'd need to use e.g. Frigate for that.
I have been trying to tackle this type of "Feature" but object detection and action detection seem to be a totally different problem. Use case: I want to "detect" when a car does not stop at a stop sign. I have researched this over youtube, reddit, etc and other than training it myself there are no models already out there, including YOLO. Can anybody offer advice on how to achieve this use case?
Try building up a method iteratively. Start by calculating the speed of a car as it crosses the camera frame.
Then try calculating the speed between two points (in car length in front of and a car length behind the stop sign).
Then set a threshold for how fast is too fast for a car to realistically go between those two points without stopping. Get notified with a video snippet when a car is above this threshold. Adjust the threshold based on the videos you are capturing.
It won’t work if your object detection is not running at your camera framerate.
My reolink cameras indicate on the clip the length of time the object in motion was recorded. One can probably figure out a way to bin these clips according to this length of time as a proxy for speeding through the stopsign.
User of Frigate here. Seems these are some pretty major differences of what you can do for free with Frigate, but if you use Clearcam, you need to pay for "Clearcam Premium":
- View your live camera feeds remotely.
- Receive notifications on events (objects/people detected).
- View event clips remotely.
- End-to-end encryption on all data.
What neither of the solutions seem to have, is encryption at rest. But I guess others, just like me, rather encrypt the volume/storage itself, instead of leaving it up to applications anyways, so might/might not matter for you.
Where can I find the list of supported GPUs? Frigate been able to handle everything I've tried so far, all from Nvidia and AMD GPUs to even Intel iGPUs.
Maybe my view of frigate and tensorflow (assuming frigate still uses it) is outdated then. I’m referring to tinygrad vs tensorflow when I say GPU support, of course google’s tensorflow is best for google’s TPUs. I’ve had better luck using tinygrad on my personal devices, but I am biased as it’s been a while since I’ve used tensorflow
This would be a good point of differentiation to make on your GitHub page or for a technical audience on your website. Frigate is SOTA in many folks minds, and to show that you are using tinygrad over tensorflow may be a good “modern-ness” signal for that audience.
Edit: another solution in this space shows a list of supported ML runtimes, which would be good info for folks wanting to run on specific hardware. https://github.com/boquila/boquilahub
Supported runtimes list would be nice, but I don't have access to much hardware to test on. I aim to remove most dependencies and support anything that can run tinygrad + ffmpeg
Unless you’re making changes, isn’t it enough to just link back to the original repo?
That said, I’ve also been in the camp that avoids AGPL-except maybe as a way to sell a commercial license while still being "open source," or just to be obnoxious. And honestly, I am still failing to see the upside in being obnoxious for its own sake.
This just seems like an extremely inconvenient, very hands-on subscription given that similarly priced AI detection exists with reliable, cheap cameras.
Closed circuit television (CCTV) is a term to describe video transmission that is not broadcast. Traditionally with BNC cables going to a control room, monitors and recorders.
I think this software-only post is meant for IP cameras / surveillance cameras. Internet is the oposite of closed circuit.
Maybe CCTV is used as a synonym for surveillance now in some regions of the world, but certainly confusing for a non-native speaker.
> I think this software-only post is meant for IP cameras / surveillance cameras. Internet is the oposite of closed circuit.
I think in this case, IP is referring to IP from TCP/IP, meaning "The Internet Protocol", not necessarily over/through "public internet links", so as long as you're only within your own local network/WAN, wouldn't that still be CCTV then? Or maybe the "closed circuit" thing is more of a physical property than I read it to be?
It's even recommended when building out a CCTV system with cheap Chinese IP cameras that like to phone home all the time. Stick 'em on a VLAN which can't access anything besides your local NVR.
I suspect, since this is HN, you're looking for something more open and hackable. But you might want to check out SIG Connect if you just want something that works out of the box.
Sorry I can't get into "how it works". Feel free to DM me for any questions.
I've been trying to use the latest versions of LLMs to analyze a target for scoring, but it's completely outside their current capabilities and hasn't worked.
Additionally, the app doesn't seem to be available in Germany.
Also check out frigate (https://github.com/blakeblackshear/frigate)
I love frigate... just had my neighbors come by because their dog was sick and were wondering if the dog had got into something in our back yard. Pulled up frigate, searched for "black dog" on the backyard camera, and found all the video of their dog.
dogs eat "post-processed food".
I wonder if there's a market for an AI dog collar that says "FIDO! don't eat that!"
Also discussed here on HN greatly:
https://news.ycombinator.com/item?id=44794508
So the app is free to download from the Apple site, and will run free, and is open source, but you have in-app purchases, and certain features can’t be used until you pay for them, is that right?
What are the paid features and what are the costs? Do I have to install the app to see the list of paid features and costs?
You might get a better response from HN if you give us more info up front.
Paid features are Live and event clip viewing over the internet, and receiving iOS notifications. You're paying for use of my server in those cases though, not for features I've made closed source. You can edit the code to use your own server if you wish too.
I'm new to HN and thought shilling the paid stuff violates the rules, so I didn't mention them.
(I'm a mod here) - it's fine to talk about paid features, as long as it's clear which ones are paid and which ones not.
The only thing that wouldn't be fine is to post a Show HN with no way to try the product out (https://news.ycombinator.com/showhn.html) and you're fine on that part.
"I'm new to HN and thought shilling the paid stuff violates the rules, so I didn't mention them."
HN ain't a non profit charity, but is the forum of a venture capitalist company, so talking about paid things does not violate any rules.
Paying for things does cause some folks to champ at the bit, though, so his assessment is not unwarranted.
It mostly gets people going on and on about how they could do it themselves for free.
Am I reading your README correct, that in order to sign up to use the app on Android, you have to install and sign up using an iOS device (using Apple's payment system) and then login on Android using the credentials you created?
Yeah sorry that’s confusing, I need to change or remove it until I’ve a payment system setup.
There is an unfinished but functional APK and android project in the repo, but it’s not on the Google Play store yet, their approval process for new individual devs is long
Anyone can recommend a good quality camera without spyware and ideally open sw stack. I am willing to do it myself with little soldering etc. that’s one rabbit hole didn’t enter yet
Depending on your definition of "good quality", you might find this project useful: https://thingino.com/
Most cameras on that list are low cost, typically with 4-5MP sensors. They don't compete on the high end in terms of image quality but you will have an open source firmware stack with root access over SSH.
Models from Eufy, Cinnado, Jooan, TP-Link, WUUK, Galayou are relatively easy to source on Amazon or Aliexpress.
The best option is just to assume any IPCam is unsafe and firewall them off in my experience; even with a fully open source camera stack connecting it directly to internet is not that great a practice. Put them on a no internet access VLAN and you can largely buy whatever cheap IPCams you want, etc etc. If you want remote access you should expose the server running the camera management software/NVR securely, not the cams.
This is basically how I run Frigate at home today, with only the NVR able to reach the camera IPs on my no web access “internet of nothing” VLAN.
Related to the original comment - can anybody recommend a budget router with vlan?
Buy recycled last gen corporate hardware off ebay or similar.
What about the Ubiquity gear? It’s maybe not AS open as you would prefer but no spyware and required cloud services is a big win in my book.
It's not open source but used Axis cameras are pretty cheap and have rtsp and onvif support. Those mostly come from commercial installs and can be configured offline using a web interface.
Axis cameras are great. Their product support is awful.
For used cameras I don't expect to get any form of official support. IMO their documentation is clear and they provide software updates for 7 years.
Nothing good has an open software stack. There are some brands (eg: Axis, Bosch, Hanwha), that support 3rd party apps that can run on the camera and perform various tasks, including AI applications.
Any product that would fall under the good quality segment is primarily targeted at the commercial market, and nobody there is looking for open software.
There's https://openipc.org/ , if open source camera firmware is of interest to you. I actually ordered a few supported IP camera modules (basically complete IP cameras but without the case) from Aliexpress and tested that I'm able to compile a firmware, I shall see if I get it working once they arrive.
It's not quite clear to me what the firmware is actually able to do, though. Apparently its motion detection is very basic, though, so you'd need to use e.g. Frigate for that.
reolink is acceptable
Check out frigate for a full NVR solution, including object detection
https://frigate.video/
I have been trying to tackle this type of "Feature" but object detection and action detection seem to be a totally different problem. Use case: I want to "detect" when a car does not stop at a stop sign. I have researched this over youtube, reddit, etc and other than training it myself there are no models already out there, including YOLO. Can anybody offer advice on how to achieve this use case?
Try building up a method iteratively. Start by calculating the speed of a car as it crosses the camera frame.
Then try calculating the speed between two points (in car length in front of and a car length behind the stop sign).
Then set a threshold for how fast is too fast for a car to realistically go between those two points without stopping. Get notified with a video snippet when a car is above this threshold. Adjust the threshold based on the videos you are capturing.
It won’t work if your object detection is not running at your camera framerate.
With ByteTrack or other similar algorithms I don't think it would be too difficult. If you're looking to contract the work out I could give it a go.
My reolink cameras indicate on the clip the length of time the object in motion was recorded. One can probably figure out a way to bin these clips according to this length of time as a proxy for speeding through the stopsign.
BoquilaHUB also does this, but with Rust: https://github.com/boquila/boquilahub/
how does this compare with frigate?
User of Frigate here. Seems these are some pretty major differences of what you can do for free with Frigate, but if you use Clearcam, you need to pay for "Clearcam Premium":
- View your live camera feeds remotely.
- Receive notifications on events (objects/people detected).
- View event clips remotely.
- End-to-end encryption on all data.
What neither of the solutions seem to have, is encryption at rest. But I guess others, just like me, rather encrypt the volume/storage itself, instead of leaving it up to applications anyways, so might/might not matter for you.
The author states elsewhere that the payments are for the use of their server, which can be reconfigured.
Yeah, I'll admit to not having tried Clearcam myself, I was just going by the information from the README as-is.
fewer features, easier setup, with more GPUs supported. (I've not used frigate myself though, only watched videos)
Where can I find the list of supported GPUs? Frigate been able to handle everything I've tried so far, all from Nvidia and AMD GPUs to even Intel iGPUs.
same here -- it's also among one of the only things to support Coral devices and RPi video cores.
I would imagine any GPGPU compute-capable pre-CUDA thing probably won't cut it.
I have used Frigate for years, I think early on it didn't support all of those GPUs. So it might be that said videos are out of date.
Maybe my view of frigate and tensorflow (assuming frigate still uses it) is outdated then. I’m referring to tinygrad vs tensorflow when I say GPU support, of course google’s tensorflow is best for google’s TPUs. I’ve had better luck using tinygrad on my personal devices, but I am biased as it’s been a while since I’ve used tensorflow
This would be a good point of differentiation to make on your GitHub page or for a technical audience on your website. Frigate is SOTA in many folks minds, and to show that you are using tinygrad over tensorflow may be a good “modern-ness” signal for that audience.
Edit: another solution in this space shows a list of supported ML runtimes, which would be good info for folks wanting to run on specific hardware. https://github.com/boquila/boquilahub
Supported runtimes list would be nice, but I don't have access to much hardware to test on. I aim to remove most dependencies and support anything that can run tinygrad + ffmpeg
Sorry, which one are you talking about, frigate or clear cam?
it's AGPL so you have to give anyone that views your camera feeds a copy of the source
Unless you’re making changes, isn’t it enough to just link back to the original repo?
That said, I’ve also been in the camp that avoids AGPL-except maybe as a way to sell a commercial license while still being "open source," or just to be obnoxious. And honestly, I am still failing to see the upside in being obnoxious for its own sake.
It's AGPL because Ultralytics requires it to use YOLO: https://github.com/blakeblackshear/frigate/pull/10717
I'd make it MIT tomorrow if you know a workaround or alternative model
Thanks!
Both for the explanation and for being a person that respects the license even if you (like me) don't particularly like it.
I see this is clearcam. What is clearam? The readme mentions it a few times so it's probably not a typo.
typo I've just fixed lol, thanks for the heads up
This just seems like an extremely inconvenient, very hands-on subscription given that similarly priced AI detection exists with reliable, cheap cameras.
How come you didn't write the iOS version in Swift?
Do we still call it CCTV if it's an IP network?
Closed circuit television (CCTV) is a term to describe video transmission that is not broadcast. Traditionally with BNC cables going to a control room, monitors and recorders.
I think this software-only post is meant for IP cameras / surveillance cameras. Internet is the oposite of closed circuit.
Maybe CCTV is used as a synonym for surveillance now in some regions of the world, but certainly confusing for a non-native speaker.
> I think this software-only post is meant for IP cameras / surveillance cameras. Internet is the oposite of closed circuit.
I think in this case, IP is referring to IP from TCP/IP, meaning "The Internet Protocol", not necessarily over/through "public internet links", so as long as you're only within your own local network/WAN, wouldn't that still be CCTV then? Or maybe the "closed circuit" thing is more of a physical property than I read it to be?
I'm also non-native English speaker FWIW.
You can use IP on a LAN with no outside access.
It's even recommended when building out a CCTV system with cheap Chinese IP cameras that like to phone home all the time. Stick 'em on a VLAN which can't access anything besides your local NVR.
“CCTV” has better optics than “surveillance camera”.
Better as in better lenses?
Or better look (as in spyware vs CCTV). Curious, too!
Yes.
Makes sense. :)
No.
This is sick! Thanks for open sourcing it!
Are there models that accurately register bullet impacts and calculate scores on shooting targets?
DISCLAIMER: I work on the SIG Connect app.
SIG Connect does exactly that: https://www.sigsauer.com/sigconnect
I suspect, since this is HN, you're looking for something more open and hackable. But you might want to check out SIG Connect if you just want something that works out of the box.
Sorry I can't get into "how it works". Feel free to DM me for any questions.
I've been trying to use the latest versions of LLMs to analyze a target for scoring, but it's completely outside their current capabilities and hasn't worked. Additionally, the app doesn't seem to be available in Germany.
How can I get in touch with you?
Just added my email to my profile. Feel free to shoot me an email!