Frigate #2
-
|
I know from a convo at frigate discussions that your team is working on support for memryx. I just bought one off amazon that I should have in a couple weeks it seems. Can I get access to the repo where your support is so I can test as well? I want to see improvements over my nvidia 1660 super |
Beta Was this translation helpful? Give feedback.
Replies: 12 comments 55 replies
-
Beta Was this translation helpful? Give feedback.
-
|
Hi @JoshuaSeidel, Thank you for giving it a try and for your interest in our work! We are currently developing a fork of Frigate with Memryx support, and it's under active review. While the project is still in its early stages and may have some rough edges or performance issues, you're welcome to test it out and share your feedback. You can find the repository here: Frigate_MemryX Getting Started
Please let us know if you encounter any issues or have suggestions for improvements. |
Beta Was this translation helpful? Give feedback.
-
|
@tim-memryx @mzidan-memryx i got drivers installed and i see /dev/memx0 but i get the below in frigate
|
Beta Was this translation helpful? Give feedback.
-
|
The issue above might be due to Frigate failing to initialize the MemryX detector because of an error connecting to the First, on the host, make sure the mxa-manager process is listening for connections: Then make sure your firewall allows connections from inside the container to the host. For example, if you're on Ubuntu and use After making these changes, restart the mxa-manager service on the host:
Please let us know if this solves your issue. Thank you! |
Beta Was this translation helpful? Give feedback.
-
|
im using the unraid system in the other discussion on this repo. what can i do to test to see if things are running? |
Beta Was this translation helpful? Give feedback.
-
|
@mzidan-memryx I haven't fully looked into the customised Frigate container but would it be possible to make the Address/DNS and port customisable in the Frigate container so that users can specify a IP if DNS is not working properly where I'm not really a big friend of exposing the whole network gateway from the host with: Sorry if that's already possible but I haven't found a way to do that. |
Beta Was this translation helpful? Give feedback.
-
|
@ich777 ill send some screenshots in a few, I don't think it's complicated at all. Some of my containers need to be on certain vlans. This allows for extremely easy configuration. |
Beta Was this translation helpful? Give feedback.
-
|
Troubleshooting VLANs and networking in general is very complicated, I tired this a few times on the Unraid forums and it never went well. However I'll try the container on a vanilla Unraid setup when the module arrives. |
Beta Was this translation helpful? Give feedback.
-
|
everything else works, so i would expect something as simple as being able to access a daemon. this is the only thing that doesnt work. |
Beta Was this translation helpful? Give feedback.
-
correction, i do not have host connected to custom networks. i just saw the setting your talking about. i did try enabling but it made no difference |
Beta Was this translation helpful? Give feedback.
-
@tim-memryx if your using local mode, then why go through all of this trying to connect to the IP of the host? and just have the device connect through /dev/memx0? |
Beta Was this translation helpful? Give feedback.
-
|
So i was able to pass in the IP of my host, but its still not working |
Beta Was this translation helpful? Give feedback.
-
|
@tim-memryx @ich777 @mzidan-memryx even adding a network to the container for the same network as the host and pointing to the right IP i still get same error
the service is running and exposed to 0.0.0.0:10000 |
Beta Was this translation helpful? Give feedback.
-
|
So I received my module today and I did a lot of testing and I'm happy to report it works (at least with a bit of fiddling around)! :) First thing that I had to do is to build the container from scratch, didn't notice that it wasn't already prebuilt, I assume you plan to create a PR so that it can be integrated upstream correct? Would it be maybe possible to build the container if you push changes at the beginning of the file and replaced this line with: After that the container started but I got this error: @tim-memryx & @mzidan-memryx I found out that if you want to run the container without I've even made a small goofy test with a car picture on my mobile phone and it is working even if it's barely visible on camera: and here it detects my iPhone, even in pretty bad lighting conditions: @JoshuaSeidel I actually saw that I had an error in the plugin and it wasn't listening on EDIT: Forgot my Please note that this is all very Unraid specific and I removed my registry where I host this container because of privacy reasons. |
Beta Was this translation helpful? Give feedback.
-
|
I think the main issue is that Frigate_MemryX doesn't allow specifying custom IP addresses since with that it would even work in your case because I assume you can ping the Unraid host from the Frigate container. You can workaround with iptables but this will get complicated quick and is just a dirty workaround and I'm not even sure what would be necessary in your case. EDIT: Please see: #5 |
Beta Was this translation helpful? Give feedback.
-
@mzidan-memryx @tim-memryx @abinila4 is this a change your willing to make in your implementation, to support changing the IP in your config? This will allow a lot more capability |
Beta Was this translation helpful? Give feedback.
-
|
@ich777 i found your fork, and the branch, built the container, and ran it pointing to my unraid ip, keeping the container at the secure vlan. and everything works. simple fix to be able to override the IP. @tim-memryx PLEASE implement this small change. so when frigate merges this in, MemryX will be the only detector i use |
Beta Was this translation helpful? Give feedback.
-
|
extra points if you can make it support the yolonas 640x640 frigate+ models that comes out. they to date support onnx. which from my understanding onnx models should work with your device nativly @mzidan-memryx @tim-memryx |
Beta Was this translation helpful? Give feedback.
-
@JoshuaSeidel, absolutely we’ll add an IP override option. |
Beta Was this translation helpful? Give feedback.
-
|
Im running this today in my production instance to see how memryx performs. ith @ich777 changes it connects. infrance speed is a little bit slower than my nvidia 1660 super. it wont take the yolonas 320 320 from frigate+. was giving labels/opject list errors |
Beta Was this translation helpful? Give feedback.
-
|
Hi @digiblur, This error indicates that the Please verify the following:
If the service is not active or the mount is missing, MemryX will fail to initialize. Please verify these and let me know — I can help you further from there. |
Beta Was this translation helpful? Give feedback.
-
|
@digiblur are you on Unraid, your username seems familiar. Can you please execute @JoshuaSeidel already made an Issue on GitHub that it's not working but never responded so I couldn't troubleshoot further. If possible please create post on the forums with your Diagnostics so that I can see what's going on, however if you are not on Unraid, please ignore my post. @abinila4 is it possible to close/lock such "Discussion" threads since it's really hard to follow because you can make posts to already answered posts in between and everywhere. |
Beta Was this translation helpful? Give feedback.
-
|
@ich777, Sure, I’ll look into locking this discussion and will probably work on adding a community page for Frigate for troubleshooting. Thank you. |
Beta Was this translation helpful? Give feedback.
-
Yup.. you helped with the Coral driver way back. As always thank you for all the driver plugins! This is the output |
Beta Was this translation helpful? Give feedback.
-
Yes, that mount is in my container template and I also did a manual docker run setup with it to no avail. I can see the mount inside the container without issue. systemctl results in command not found on my box. |
Beta Was this translation helpful? Give feedback.
-
|
@tim-memryx is it planned for models with more labels like packages, ups, FedEx, usps, Amazon, dhl etc, to make it into the frigate PR? And also are you guys trying to get the PR merged prior to 0.16 release? I see everything's approved just some conflicts now due to development since PR was entered. Really eager to make frigate with memryx my daily driver. (Happy to stay beta tester as well if upstream changes stay synced) |
Beta Was this translation helpful? Give feedback.
-
|
Hi @JoshuaSeidel, Thank you so much for offering to beta-test the upstream Frigate updates—we really appreciate it, and we’re excited you’re considering the MemryX accelerator as your daily Frigate driver! As you noted, the PR is nearly complete. Our team is currently focused on shipping a new MemryX SDK release, but once that’s wrapped up, we expect the merge to move forward quickly. Regarding package-detection models, our hardware and software are fully capable of supporting them. Since this functionality is part of Frigate+, we first need to coordinate with the Frigate team to integrate it, and we’ll keep you posted on our progress. Thanks again for helping us improve Frigate integration! |
Beta Was this translation helpful? Give feedback.
-
|
@mzidan-memryx i know your device is. The models that are included don't have the label in the label maps. I tried running frigate+ on 320x320 onnx but it wouldn't load since I read that mx3 can use onnx models |
Beta Was this translation helpful? Give feedback.
-
|
Hi @JoshuaSeidel , If your custom model is in ONNX format, we do support it—but you'll need to compile it into our MX3 .dfp (Data Flow Program) format before running it on the device. You can follow the instructions in our Neural Compiler documentation to compile ONNX models for MX3. That said, please note that the current post-processing code in Frigate is written specifically for 80-class models (e.g., COCO). So, you might encounter label-related issues if your label map differs. Apologies for the inconvenience—feel free to reach out if you need help compiling the model. Also, we may need to make some updates to the YOLO-NAS post-processing logic to fully support custom class configurations. Thank you! |
Beta Was this translation helpful? Give feedback.
-
|
@abinila4 i'll take a look to see if i can find the frigate+ model locally that i can pump through your program to make it compatible. i was just hoping to be able to jump right into using my MX3, package detection in frigate is something big i use. |
Beta Was this translation helpful? Give feedback.
-
|
@mzidan-memryx @abinila4 is there plans to release a arm64 macos sdk? |
Beta Was this translation helpful? Give feedback.
-
|
I managed to recompile my Frigate+ model into .DFP with Claude's help. Not particularly easy and I have a request to the Frigate+ team that they add .DFP compilation into the available Frigate+ model downloads. But right now, mx3 support in Frigate seems to be a bit alpha-ish so I wouldn't be surprised if they're waiting for the dust to settle a bit. I hope Frigate does add MX3 for first class support as we really need a lower cost option than Haile-8. |
Beta Was this translation helpful? Give feedback.
-
|
Hi All. Not much action here for a fair while. I know MX3 support is out in .17 + and I was testing all the .17 beta releases, but now my MX3 has just arrived from the US, and I'm looking for frigate config ideas in regard to processing options. i.e. Which processor to assign to different Frigate tasks. My Frigate machine is a dedicated mini PC (no hypervisor) running an Intel® Core™ Ultra 5 Processor 125H,14C/18T with Intel® Arc™ graphics, 32GB RAM and a fast M.2 SSD. The current config of 12 cameras is fast using the CPU, GPU and NPU. I have not yet tried LPR and Face Rec' yet but am going to soon. Any recommendations for where to best utilise the MX3 in this setup? |
Beta Was this translation helpful? Give feedback.
-
|
Hi @CMGFrigate, Great to hear your MX3 has arrived. For your setup, we’d recommend using the MX3 as the Frigate detector. You can install the MX3 in the M.2 slot and configure Frigate to use it as the detector hardware. In the config, you can set it up like this: detectors:
memx0:
type: memryx
device: PCIe:0This allows the MX3 to handle detection while the host CPU/GPU remains available for other tasks. Other Frigate tasks, such as video decode, streaming, LPR, and Face Recognition, are usually handled automatically by Frigate based on the available system resources. For available settings, refer to Frigate’s configuration You can refer to our setup guide here for configuring Frigate with different model options: We also have a Frigate community page here: Please feel free to reach out with any further questions. Thank you. |
Beta Was this translation helpful? Give feedback.
Hi @JoshuaSeidel, Thank you for giving it a try and for your interest in our work!
We are currently developing a fork of Frigate with Memryx support, and it's under active review. While the project is still in its early stages and may have some rough edges or performance issues, you're welcome to test it out and share your feedback.
You can find the repository here: Frigate_MemryX
Getting Started
Install MemryX Locally
Before running the Docker setup, you need to install the latest MemryX on your local machine. You can do this by running
./user_installation.shinside thedocker/memryx/directory later.Run the Docker Setup
Execute the startup script:
Update the Conf…