Yes.
Here is the link to the version I am using.
Also, this is the screen shot.
1 Thank
why its R9 precision is so bad?
did you record Ra measurements? How about Ra precision?
Please let me know why. You are the developer.
Please read. Ra numbers are in my previous post. Why are you asking obvious questions when the answers are right in front of your eyes??
sorry
seems ML method is not that bad in terms of Ra/R9, but I think spectral reconstruction method’s generalization ability is much better than ML method, you can test other light sources, for example CFL light and daylight, I can guarantee that ML method will fail, especially for CRI(Android/iOS app uses ML method, while my PC software uses spectral method).
1 Thank
Thanks again for all the time-saving links! Still no success at loading Opple Home (data analysis error) but was able to load Windows desktop and will give it a try this morning. Just when I thought I was done, LM4 sucked me in! 
Hi Steve, does this statement applied to Lightmaster 3 as well? In other words, LM3 would be “better” if someone were to develop Windows app for it using spectral method?
Try and find the modified Android apps that were posted in here. They tweaked the package so that it should no longer require the items giving you errors.
1 Thank
it’s complicated to achieve spectral calibration. we don’t have a device named monochromator, it’s very specialized and professional, average company does not possess this sophisticated device. plus Light Master 3’s sensor is no longer available on the market now, so LM3 is dumped.
Hi @Limsup and @stevechang - I tested the desktop Windows app for LM4 with about 25 lights today. Thanks again and allow me to share some observations. My test was done with light on tripod and hotspot on the sensor. I think I was about as compulsive and careful as possible considering this is amateur-level testing. Strange outliers were repeated to make sure it wasn’t me who is “outlying.”
Strictly with respect to R9, I am sorry to say for me the Windows app failed where it hurts - the group with the highest CRI’s. And it failed both ways, high’s reported low, and low’s reported high. In other words, there is no trend, no pattern; just seemingly random error.
Nichia B35AM and FFL351A Rosy, two LEDs with among the highest recorded R9 (upper 90’s) measured 84 (for B35AM) and inexplicably 68 for the FFL351A rosy. E21a another high flyer, recorded 81. What is so strange is the iOS App Opple Smart passed all of these LED’s with R9 in the 90’s.
Conversely, Getian GT-FC40, with R9 in the 70’s per Simon’s test, showed 98 in M21B and 97 in M21E. Again, the iOs Opple Smart passed this with much more reasonable numbers.
Ra is mostly good, with occasional some random weirdness, like Ra 90 for FFL351a. Other than that, CCT is great, Duv is good enough for an amateur device.
The good news is that IMHO/IM brief experience, LM4 with Opple Smart is decent to good for CCT, for Ra, R9, and for Duv. With the caveat that users should test known LEDs, 219b 4500k should be minus 0.0050 to 0.0100, and 519a 4500k should be around Duv 0, etc. and observe the bias for Duv. I will re-test the weird numbers again to make sure I didn’t make mistake, but for me, Opple Smart seems to be the “best” app for LM4.
strictly speaking the sensor we used is not suitable for R9 or each individual R values measuring. since we can’t get the real spectra we are measured, the spectra shown on the windows software actually is calculated spectra not measured spectra, hence the error for individual Rs may differ from the real Rs. for Ra I think in 99% cases it’s quite accurate, because it’s averaged. Opple Smart uses machine learning, actually it’s a statistical method, it does not have much theoretical backing.
From the information given, I would say that it appears that Opple Home 3.3.1 may give more accurate R9 measurements, and from the information posted upthread by Koef3 and others, it appears highly likely that the Desktop version gives more accurate Ra measurements.
But while I don’t think it is unreasonable to trust Home 3.3.1’s R9 value slightly more, the data presented don’t seem to give a high degree of confidence one way or the other. Timing of purchase isn’t a very good guarantee your emitters are from the same tint bins as the references. Even if they are within the same bin, various measurements posted over the years by individuals with good gear seem to suggest that individual LED’s don’t always fall within the labeled bin.
To illustrate with the 5000K LH351D, for example, Zeroair’s measurement of R9 = 43 seems low in general for a ~90 CRI emitter, but not unreasonably so. The even lower measurement from the Opple Home 3.3.1 of only R9 = 30.1 would make me scratch my head a bit.
At the same time, there have been samples of the 5000K LH351D where R9 measured close to the 61.5 you measured with the desktop program. Here Maukka had a sample with R9 = 61 to 70 depending how hard he drove it.
With that said, I think it even makes a degree of sense that spectral reconstruction could compromise on the accuracy of saturated color measurements like R9 for the sake of reconstructing a spectrum that is reasonably useful for calculating Ra (assuming the actual spectrum is at least reasonably continuous like an LED, not spiky like a fluorescent light). Trying to fit a curve through 8 points could result in the curve having relatively large local errors in places. If the reconstruction specifically tries to mimic the shape of the black body reference curves, which extend well into the infrared, I could definitely imagine an otherwise very reasonable reconstruction tending to give elevated values in the red range, and therefore report R9 as higher than it really is.
In contrast, training a machine learning algorithm with the data from a sensor that has channels at 630nm and 680nm - both well within the R9 reflection spectrum - could quite credibly converge on a tuning that appropriately weights the values from those channels to produce relatively accurate R9 scores.
In any case, we always have to be careful to remember that the LM3 and LM4 are both built around hardware with significant fundamental limitations, and we’ll have to accept a relatively significant degree of uncertainty in exchange for the low price.
2 Thanks
I installed Opple Smart from the play store, but could not get past the Welcome to Opple Home/Get Verification Code page. I entered my email but got the message “Data Analysis Error32001”.
I understand there is an additional app or revision to install that gets me pass this block. Do you (or anyone) have the link to that please?
Here is the link to the apk files of the reverse-engineered app.
I think you have to allow unverified apps to be installed.
I just received my Opple 4, and I took some measurements of my ~4000K lights.
I bought it mainly for DUV measurements, and so far, the results are somewhat perplexing.
The DUV of my 519a Lanapple, which is visibly more rosy to the eye, is higher than my greener D3AA SST20, TS10, and similar to the very green IF25A SST20 and M21E FC40.
The IF25A is the most green-tinted light I own, yet its DUV is higher than the 519a by only 0.0003.
duv is relatively easier to calculate, but its precision can’t be guaranteed, it’s provided by another colleague, I have my own method, to be precise, my own coefficient. but in order to be consistent with the duv values on Android/iOS platforms, I still use my colleague’s coefficient.
in the future, perhaps I’d provide my version for you guys to test.
and keep in mind, it’s inherently difficult even impossible for each metric to be as precise as integrating sphere, after all as7341 has only 8 VIS channels.
1 Thank
I had high expectations for the LH351D in the SP10 Pro, especially since it was marketed as having high CRI. However, I was surprised by how greenish or pale it made my skin appear.
Here’s a photo I took two years ago using the same light. If you look at the top row of the photo, you can get a sense of the red content in the light and how it compares to the LatticePower 6000K and the Nichia 519A. From my observation, the 519A has the most red, followed by the TS10 (LatticePower 6000K), and then the LH351D. And this observation aligns very well with the measurements from Opple Home 3.3.1. (R9 from 3.3.1: Lattiepower 6000K=56.6, LH351D 5000K=30.1, 519A=91.6. The first two numbers are from my previous post and the number fro R9 are taken today)
The above photo is from my post here:
I’m also puzzled by the R9 value of 30.1, but it’s just a rough estimate from a relatively basic sensor. A higher number might make more sense. However, since these measurements are only approximate, the absolute value doesn’t hold much significance and shouldn’t be taken at face value. I’m using it more as a comparative tool to evaluate my own measurements objectively. In that sense, the measurements from version 3.3.1 were the most useful for me, as they provided a reliable way to measure what I see.
1 Thank
Got it this time. Thanks.