Announcing v1.0.0 of MMM-Faces.

There’s a few facial recognition modules for Magic Mirror. I’ve tried most of them, and found them a bit painful to setup. Furthermore, the setup is something that needs to be done again and again and building and capturing the images seemed awkward. I felt that there must be an easier way of managing this, so decided to build one, and then got carried away! This module is a simple wrapper around running a facial recognition system, and then sending data from that into MMM-ProfileSwitcher. All the actual work of changing the mirror config is left to MMM-ProfileSwitcher.

The above image shows the admin interface that allows you to update the recognition parameters, rebuild the model, without having to restart your mirror in any way. Lots more documentation is all contained within the README of the actual project.

With a Pi 3B+ and a Movidius compute stick from Intel, I’m running my facial recognition at about 6 frames per second and it works great, without overheating the Pi, with almost instant switching of the mirror display.

A “soft-announce” for something I’m working on, but is almost ready to be released…

I’ve had problems with all the versions of face-recognition, profile-switching, debugging… so I’m attempting to create a new module that takes out most of the pain. The pain of getting the profiles configured correctly is still there (that’s job#2), but the facial recognition… that was something I wanted to try and improve. So, pre-announcing a new face recognition module (because we all need another module doing the same thing).

https://github.com/njwilliams/MMM-Faces

I welcome thoughts/suggestions to improve this!

@dresch360 It is still supported. I wasn’t watching the forums for questions, sorry. Just “woke up” 😀. Pushed a new release that has more debug info in case it helps. Let me know what errors you get or what config you’re using if it’s not working for you.

Description:

This will display (and update on a regular basis) a random Kanji character, with information about that character including Kunyomi, Onyomi, english translation(s). The set from which the kanji is chosen has a few filters including JLPT N1-5 and elementary grade levels. This is an initial release- I would like to add in more features over time. Collaborators and ideas welcome!

Screenshots:

Download:

https://github.com/njwilliams/MMM-Kanji-Random

---

Version 1.0.0

• Initial Release

Just created v1 MMM-Kanji-Random that does this job. No stroke order yet. It’s a start!

Description

Hi Mirror Folk!

This module combines a few different pieces of data about remote places to give a handy one spot to lookup: time, weather and currency. I’m still very much tinkering with how it displays the info, but it’s at the point that I’m using it all the time so thought I’d ask for some feedback or suggestions around styling, since that’s definitely not my forte!

Screenshot

Download

[card:njwilliams/MMM-PlaceInfo]

@Mykle1 I realize it’s not the most … informative… of modules, but I was playing around with the idea of combining it with face recognition, to show the freakish eye when it’s someone it doesn’t recognize :).

This was with the new electron version and the GL drivers. I had some of the other eye candy URLs on the screen for a day or so, playing around and no issue. However, just that particular selection it would crash in about 10 minutes. Weird. Possibly the memory usage was high on the others - I didn’t do any debug until it broke, of course.

I’m happy to play around with more debugging.

New York apartment, so hallway=kitchen! I’ve got a mirror in the entry way so that I can glance at it on the way out and find out at a glance what’s going on. A couple of months ago, the mirror died (FS corruption on the SD card, which I’ve since eliminated) and it took me a while to get around to fixing - it was weird looking at the mirror when it was dead, and not seeing information. We’ve really become reliant on this!

Anyhow, I’ve put up a couple of blog entries about setting up the mirror and configuring diskless booting.

I’ve upgraded to 2.4.1 and using the suggested OpenGL driver and all is well. Decided to play with some new modules and enabled the MMM-EyeCandy and it also worked well… for a while. I found that with a particular image selected (37), it would work, but go a little glitchy - stuttering on animations. And then after something like 10 minutes the screen would freeze. And then a while later it might recover, or it might just switch to a black screen. Weird. With a bit of debug, I found that the electron process that was running the animation would consume all the memory until eventually the OOM killer would kill it. At which point, the X server would just show nothing until magic mirror was restarted. I don’t know if this is a previous known problem with eye candy, a problem with using the OpenGL driver, or 2.4.1 of MM with it’s usage of updated electron.

In terms of logged errors, there’s obviously the oom killer log showing that it killed electron. There are logged errors in the MM logs, but they are from a different electron processes to the one that was running out of memory:

• the unhandled promise rejection errors that others have asked about
• and gles2_cmd_decoder errors.

Both of those errors were from other electron processes to the eyecandy (is there an easy way to associate electron PIDs with MMM modules? I’m using guesswork…)

Anyhow, I’ve disabled the eye candy module for now and everything works fine, but curious as to how to debug this more, or how to protect things better (e.g. can electron provide simple ulimit controls for the subprocesses? Perhaps something for pm2 configuration?)