I’d appreciate if you could verify lux/cx,cy/uv/duv/CCT if those metrics are accurate. those metrics are actually calculated by other colleague. I did a few preliminary test, haven’t found any problem. plus, I have my own algorithm for calculating those metrics too, but mine is not employed in the Android/iOS app.
the app programmer complained too, but we engineers really can’t decide this.
1 Thank
For this it is important that we use the same LEDs (because the spectrum is different with each CCT/CRI). Even flashlights from same batch and/or same manufacturer and type can differ in duv/x/y.
I think that at least for now you can just measure any light source with both the spectrometer and the LM4 and check that the values for lux/cx,cy/uv/duv/CCT are reasonably comparable
1 Thank
Ok, this is not a problem.
I also ordered some proper coating for my sphere to get valid results without calibrating (since Opple LM4 give no raw data this is crucial to compare)
What app should I use for the LM4? I heard there are two apps available which do more or less the same…
1 Thank
I’ve had better luck with Home.
1 Thank
Can you tell how big the differences in the results are?
@stevechang Are there huge differences in results between these the two apps? And if so, are there different algorithms, different developers? Do you get on well with the other colleague who calculates the metrics or is he allowed to make changes to the algorithm? Therefore, I ask whether there is any chance at all of transferring the changes to the official app in the event of a significant deviation.
you’d better to use OPPLE Home
OPPLE Home is for consumers, while OPPLE Smart Lighting is for business usage.
generally OPPLE Home is more up to date, they take caution when they decide to update OPPLE Smart Lighting, this means it is not newest in terms of algorithm. I remember long ago OPPLE Smart Lighting is way more inaccurate than OPPLE Home because it is not synced with OPPLE Home.
2 Thanks
I want to add that version 3.3.1 of the Home app seems to be better than 3.5.1, because the latter has problems with low CRI lights.
See here: Opple Light Master 4 discussion thread (new 2023 model) - #749 by Limsup
And also on Reddit: https://old.reddit.com/r/flashlight/comments/18ys0fk/opple_home_app_not_showing_r_values/
Both posts talk about version 3.5.1, released around Dec 29, 2023.
So I’d say that for now Home app version 3.3.1 is the recommended one. It was released on Oct 14, 2023 for iOS and on Dec 12, 2023 for Android.
But it’s interesting to compare it to Smart app version 3.5.1 (released on Dec 8, 2023), since it would much easier for me to modify it further.
I can compare these.
But in general it depends on which algorithm is used in these specific versions. Likely there were major changes in the algorithm from 3.3.1 to 3.5.1, don’t know why though.
Why this?
1 Thank
All Opple apps are made using React Native. But the latest versions also use Hermes that optimizes the JavaScript code, but also encrypts it so that it can’t be edited without special tools. Since you can’t un-hermes it, the only way to make adjustments is to edit the bytecode directly (using this manual).
Here’s a visual example: normal JS (though minified) - encrypted JS - decoded bytecode.
All Home apps use Hermes starting from v2.0.3. But Light Master support was added only in v2.0.7. So there’s no way to avoid the bytecode with Home. For the mods I made I was lucky to find a solution that only required to edit one line of such code.
But for the Smart app encryption started to be used only after v3.5.1. So if the measurements are fine in Smart Lighting v3.5.1, there’s a better chance to implement a feature like local reports there.
It might still be possible to add features to versions that use Hermes, but I’ll have to do some experiments first. Maybe it’ll be possible with a library called Frida. I had some success with it in the past.
1 Thank
yes, but feel astonished 3.3.1 is even better than 3.5.1.
I think more data doesn’t necessarily lead to better result. because of the the F1 plus F2 SNR problem I mentioned before.
actually LM3 and LM4 till now employ AI/ML technique, we constructed a theoretical model that takes theoretical channels value as input, however the real channels value must be converted to the theoretical space to feed to the theoretical model, so there actually is a “bridge” to connect real and theoretical channel values, the “bridge” is the key for LM3’s success. but the “bridge” does not lead to accurate results for LM4. all the updates(3.3.1, 3.5.1) is for fine-tuning the “bridge” nothing else. but due to the constraint of the inherent characteristics of the chip channels’s distribution, this problem can’t be easily solved. what you can do is to collect larger data to fine-tune the “bridge”, however this is not necessarily correct to just collecting more data.
so spectral reconstruction is due to replace AI method.
sometimes I think LM3’s whole systematic design is delicate, however it is not intuitive and its explainability is not very good. spectral reconstruction is better, and I think it would bring much better accuracy/consistency/explainability.
however everybody knows the chip(as7341) only has 8 VIS channels, while cl-500a can output from 360nm to 780nm in total 421 intensity values, so its resolution is enough for capture the minor details in a complicated spectrum, so in most cases don’t compare LM4’s accuracy with cl-500a. but I still think if we call achieve as7341’s utmost potential, that’s enough, and its accuracy will be acceptable to you guys.
5 Thanks
I think first thing is to compare Home and Smart lighting app (the modded ones so I don’t need the account for using the LM4), in conjunction with my spectrometer to get a baseline which app is finally better in calculating the values.
If it turns out Home is better, then we have to deal with the obfuscated code… we will see. 
@stevechang Would be cool do get better values from the AS7341 - the sensor is not that bad and I think we can get some accuracy from it
4 Thanks
It’s a bit off-topic. Thanks Steve for showing me around Changshu!
9 Thanks
it is my pleasure to show you around the city where I live, perhaps next time we should get prepared and explore more in-depth. plus I need to get more familiar with Led related things, I can not grasp the real essence of your demonstrated stuff actually.
4 Thanks
Maybe at some time you want to visit Germany? 
Can show you some stuff
2 Thanks
well, perhaps someday
some colleagues returned from Frankfurt last month
1 Thank
Is there such a thing as flashlight meetups? 
The custom keyboard scene regularly organizes meetups where people meet, talk about keyboards, try the keyboards others brought with them, or just chat and see the faces behind the nicknames.
Would be fun to have such things for flashlights, too, I guess, haha.
1 Thank
What’s so amazing about it? Meet up, have a live conversation. Show and compare new and old flashlights. )))