Seek Compact XR lemon or not?

[eecs]

What temperature did the camera report for the 18°C scene?

I found the photo i took during the 12 min test and the Min and Max were both 18c

Wow, so spot on at ambient, compared to an error of at least 6°C and sometimes up to 8°C on my camera!

The camera stops clicking when in the gallery, After i leave the gallery i get a lot of red around the edge of the screen then yellow and then green in the centre, the camera does 5 clicks fast and then i get some yellow full image and then green full image for maybe 10 seconds before the camera settle back to green across the full image.

OK, so a full recovery within 10 seconds, compared to my camera that still is not back to optimal performance after several minutes.

I can understand why you are happy with your one!

Now I just need to work out if you got an unusually good one, or I got an unusually bad one...

[Tinderbox (UK)]

What temperature did the camera report for the 18°C scene?

I found the photo i took during the 12 min test and the Min and Max were both 18c

Wow, so spot on at ambient, compared to an error of at least 6°C and sometimes up to 8°C on my camera!

The camera stops clicking when in the gallery, After i leave the gallery i get a lot of red around the edge of the screen then yellow and then green in the centre, the camera does 5 clicks fast and then i get some yellow full image and then green full image for maybe 10 seconds before the camera settle back to green across the full image.

OK, so a full recovery within 10 seconds, compared to my camera that still is not back to optimal performance after several minutes.

I can understand why you are happy with your one!

Now I just need to work out if you got an unusually good one, or I got an unusually bad one...

I added the image with the 18c high and 18c low to my previous post.

But remember i have the "Seek Compact with the adjustable focus not the XR version", though they should both have the same sensor.

John.

[IwuzBornanerd]

Hopefully I have all the syntax & indents correct.

Just needed to increase the allocation size for diffcurve from 13320 to 14430, otherwise OK.

Of course 

In the config file I have set the pixel 1 value to 6552, ForC to C, and ambientT to 20, but I didn't know what to do with diffshouldbe so I left it at -487.

That's as good as anything at this point.

Redoing the calibration with your new code didn't seem to make much difference. Should I calibrate with smoothing on or off, or doesn't it matter?

The settings have no effect on the calibration operation.  For calibration the program takes the values from frame 3 (your external uniform temp. scene or an average of those frames) and subtracts the values from frame 1 (the image of the flag) to get a set of what I call "diffs".  Then the program divides those diffs by the surmised value that all those diffs "should be" if the sensor were perfectly as designed with no pixel to pixel variations.  That is what diffshouldbe is.  It needs to be tweaked but something close to 500 has worked quite well so far.

As I look at my latest (non Python) code and the Python code it is apparent that I have not studied the Pro enough to have a good idea how to handle the new frame 9.  What I gave you was a workaround that I have in my latest code which is intended to get the curve numbers similar to those of the old frame 9.  Along with that it seems that I should have done something similar to the pixel 1 value but that is not in the code...EXCEPT that in my latest code I set it to 4872 for the Pro even though the pro reports something over 6100 on startup!   And the temperatures I get off the Pro are okay at least near room temp.  Hmmm.  I also have diffshouldbe at 687 for the Pro.

I also see that in the Python code according to the calculations it expects the ambientT to be in Fahrenheit, not either F or C as stated in the config.  "My bad"--forgot I was going to accommodate that I guess.

So you may get temperatures closer to real if you put ambientT in F, list pixel1atT as 4872 & diffshouldbe at 687.

I think in order to answer your question as to whether the problems you observe are the camera we need to see images with a 6C span and see if it stays stable.  The correctness of the temperature is secondary.  No? 

Sorry to put you through such a hassle with the software.   It points out a problem with hackware--the hacker has to keep up with the changes the hackee makes.

The stuff you observe is crazy.  I don't know how it could be the camera & I think it's sad if the app is doing what it appears to be doing.  And all the more strange if it does not do it for Tinderbox. Do you guys have the same version of the app??  The version I have on Androidx86 apparently is doing it with my Pro, as shown earlier.

The black dots in the laptop image are jittering around in the sort of figure-of-8 pattern you would get if you joined the dots. I tried taking a video, and the program said it was recording, but there was no video file created - probably I messed up the installation of some necessary library.

It uses openCV for writing the videos; see if you got all the "Also now requires" at the top of the Python file.

[eecs]

I also see that in the Python code according to the calculations it expects the ambientT to be in Fahrenheit, not either F or C as stated in the config.  "My bad"--forgot I was going to accommodate that I guess.

You converted it to F early on, so that was OK:

Code: [Select]

if(ForC < 0): ambientT = ambientT*1.8 + 32 #convert to Fahrenheit for use in finding ref point on curve


So you may get temperatures closer to real if you put ambientT in F, list pixel1atT as 4872 & diffshouldbe at 687.

No luck there; the temperatures were reported as over 100°C. I tried just lowering ambientT, but it was still reading high when I got to -40°C, and then the image went black!

I think in order to answer your question as to whether the problems you observe are the camera we need to see images with a 6C span and see if it stays stable.  The correctness of the temperature is secondary.  No? 

Indeed, and the temperatures reported do seem to stable (at least when I stop fiddling with the config file). What is also interesting is that with your program I can stop the program for a few seconds, then run it again, and almost immediately get images just as good as before, and certainly without the halo effect. There is also much less of the vertical striping evident with your program compared to the Seek app.

I still seem to be getting just a few pixels that are way out and pulling the lower temperature down:

so I modified your smoothing code to pick the median of the 9 adjacent pixels instead of the mean. I then see:


Here is an image with about a 6°C span:

and here is one with a 16°C span:

and the same immediately after stopping and restarting the program:

Sorry to put you through such a hassle with the software.   It points out a problem with hackware--the hacker has to keep up with the changes the hackee makes.

No need to apologise to me - I am glad to have anything at all, and greatly appreciate you having released your code.

The stuff you observe is crazy.  I don't know how it could be the camera & I think it's sad if the app is doing what it appears to be doing.  And all the more strange if it does not do it for Tinderbox. Do you guys have the same version of the app??  The version I have on Androidx86 apparently is doing it with my Pro, as shown earlier.

Since both Tinderbox and I only very recently bought the cameras, and since he is on a Samsung S3 and I am on a Samsung S5, I think it almost certain we both run the latest Play Store version. For what it is worth my one reports itself as version 2.1.5.2, with Image Processing Software 2.5.6.2.

My temperature error could be bad calibration data being supplied by the camera to the app, as could the vertical stripes and some of noise. It is hard to imagine how the same app could recover quickly from a warm start for Tinderbox but not for me, especially when your program recovers from a warm start for me with no problem!

Anyway, thank you very much for your help (you and the other contributors in this thread). I don't think it is worth you wasting more of your time on my problem; I have satisfied myself that my camera/app combination is not working as it should, so I would most likely return it even if we could get the hardware to work with a third-party program.

[IwuzBornanerd]

You converted it to F early on, so that was OK:

  You know, I thought I did that but didn't find it when I looked for it last night.

so I modified your smoothing code to pick the median of the 9 adjacent pixels instead of the mean. I then see:

Nice idea, I might try that out; but you have a wider span there wherein those pixels might not be different enough shades for you to see them.

Here is an image with about a 6°C span:

Now THAT is what you wanted to see in the beginning, right?  You can see the shape of the thing.

My temperature error could be bad calibration data being supplied by the camera to the app, as could the vertical stripes and some of noise. It is hard to imagine how the same app could recover quickly from a warm start for Tinderbox but not for me, especially when your program recovers from a warm start for me with no problem!

My guess on the temperature is that they don't have any number stored on the camera that ties the ambient temperature to a particular value reading from the temperature sensor thus exposing the reading to the full tolerance of the temperature sensing device.  That is of course just a guess.  The 2 variables I have in the config file are my attempt at eliminating that problem.

Happy you were able to prove that the camera essentially works.  Personally I don't think there is anything wrong with the camera or any data on it but as you say,  you need the app & the camera to work together.   From what I read on this forum the Seek app is their biggest anti-selling point.

[IwuzBornanerd]

But remember i have the "Seek Compact with the adjustable focus not the XR version", though they should both have the same sensor.

John.

Yes, they have the same sensor & pcb assembly number & revision--at least between a 2015 XR & an end-of 2016 non-XR

[eecs]

I did send some questions to SEEK Thermal on Monday, but so far all I have back is an automated reply saying that because their support desk is closed on Fridays they may take up to 10 business days to reply!

The next response you get might be from someone who ignored your question & said to check this, that, and another link and if you don't find an answer then reply to the message...and by the way we closed your "ticket" but we'll re-open it if you reply. 

Seek replied after only 5 business days, and said that between 0°C and 100°C the Compact series should be within ±5°C and that accuracy should improve over the first 1 minute from cold. Since my unit's accuracy deteriorated over about 5 minutes from cold, deteriorated further if stopped and restarted, and ultimately reached an error of 8°C when fully warm and after a restart, it was clearly out of spec and I have now returned it. Amazon did not give me the option of a replacement (refund was my only choice) so now I have to decide if I want to risk buying another Seek XR, save up for something like a TE Q1 pro, or wait a year or two for the technology to get better and cheaper...

[IwuzBornanerd]

 It's hard to beat that $170 (US) price for 30k thermal pixels isn't it?  My guess is that this is Amazon's close-out sale on the micro USB version.  They also have had the slow frame Pro on sale for $270 & that is why I bought one.  NOW they say it "ships in 1-2 months" so maybe Seek isn't making them anymore since phones are going to USB-C.

I also noticed that Amazon is not selling ANY "Fast Frame" Pros and Seek's feature comparison table for the compact models lists the Pro as <9Hz, so I wonder if Seek has given up selling FF units--maybe because it takes a 2GHz phone to get FF out of it.

I know you said not to spend more time on the matter but since I have 5 Seeks I thought I'd run some tests on them with respect to the issues reported in this thread to see what sort of "spread" in performance there is on the Seek app, now that I can actually run the Seek app.

Here is a summary of what I found:

Model	Start temp C	Settled temp C (3-4min)	unplug/gallery
June 2015 XR	35	26	Comes back with 18-22C pattern, keeps pattern but recovers to 23-25C on the pattern
Jan 2017 nonXR	30	25	Recovers in a few seconds
Jan 2017 XR	35	29	Recovers in a few seconds
Dec 2017 Pro FF	33	29	24-33C degree pattern, recovered to 1 degree pattern one time
Sept 2019 Pro	32	30	27-32 degree pattern, recovered after gallery but not all the time

In all cases the camera was face down on a piece of paper that was about 9 inches from the exhaust of the laptop computer running the app.  I didn't realize it until I measured the temperature of the paper with an IR thermometer, but the paper was heated by the laptop & was hotter at the edge than where the camera was and was warmer than room ambient by about 2C where the camera was.  So the paper was 27C according to the IR thermometer.

I went back to apkpure.com to get an older version of the app to see if it displayed the same problem and I found a just-released version 2.1.6.0 & tried that, but it did not fix the disconnect/gallery problem.  In fact, the temperatures were farther off with the Pro units on that version of the app.  None of the older versions of the app even worked with the camera on Android x86.

Using an oscilloscope I see that the USB communication does stop when you go into the gallery or do anything else in Android, so it effectively disconnects the camera, but I have no way of knowing if the app re-initializes the camera or re-loads the on-camera data.

I thought perhaps the pattern issue was due to the camera being stable at temperature while the software was expecting it to change so I let the FF Pro warm up, disconnected & reconnected it & then put it in the hot exhaust of from the laptop for a few  minutes and then back face down on the paper.  The pattern changed!  I think it did go back to the first pattern after cooling back down, though.

FF Pro after reconnect:


Then after heating:


Another thing that is odd is that after the Pro's warm up on the hot-off-the-press app they have a pattern (only 1C but a pattern):


What a mess. 

[eecs]

I know you said not to spend more time on the matter but since I have 5 Seeks I thought I'd run some tests on them with respect to the issues reported in this thread to see what sort of "spread" in performance there is on the Seek app, now that I can actually run the Seek app.

So all your Seeks are within the specified ±5°C after warmup (at least the first time from cold).

We can probably forgive the noise and patterns on a 1°C range.

That just leaves the extreme (10°C range on a uniform scene!!) gradient seen after briefly viewing the gallery or doing anything else that stops the camera. Do you get any sign of this happening if you do the same test using your software, or is this a "feature" of the Seek app rather than the hardware?

[IwuzBornanerd]

As I said before I have never seen such an obvious pattern when reconnecting with m y own software, but for the sake of completeness I ran some tests, since I don't regularly unplug & immediately re-start the cameras.  Here goes... 

FFPro warmed up:


FFPro after 10sec unplug (program waits 10sec to try reconnecting):


FFPro a minute after re-attach (did not take all that time to recover):


2015XR cold start:


2015XR warmed up:


2015XR after 10sec disconnect:


2015XR shortly after re-connect:


2017XR cold start:


2017XR warmed up:


2017XR after re-start (program could not reconnect after unplug):


2017XR shortly after re-connect:


So there is a bit of a pattern but it is not much of a temperature span except perhaps with the 2015XR, but that camera has 4 times as much noise as the newer ones so it's easier to look bad.  I skipped the slow frame Pro because I don't have the software coded to use a special set of NUC files for that one since I just got it and am working on making the code more versatile in that regard.  I skipped the non-XR because it tested similar to the 2017XR in the Seek app tests.

[Tinderbox (UK)]

Just tried my Seek Compact again since i updated there was a software update a couple of day`s ago.

Same test camera pointing down at my wooden floor, Room temp 17C

Started camera a pointed it at the floor off the edge of an table high temp is 19C but within 10 seconds it dropped to 17c, I also have an hand-held infra-red thermometer and it says my floor is at 17C

One problem with the new software i have to make sure the app is ended/closed not sitting in the background or it will not detect my camera when i plug it in.

John.

[IwuzBornanerd]

Just tried my Seek Compact again since i updated there was a software update a couple of day`s ago.
...
...
One problem with the new software i have to make sure the app is ended/closed not sitting in the background or it will not detect my camera when i plug it in.

John.

Version 2.1.6.0?  I checked for that problem here.  I don't see it under Androidx86.  Plug in camera, app pops up asking if it's okay to access the device; run app; disconnect camera; open terminal emulator or file manager, plug in camera & app pops back up.  No need to close out of app.

Crazy app, crazy OS; happy I don't need to use either. 

[Tinderbox (UK)]

Software version 2.1.6.0, Image Processing Software 2.6.1.12

My Galaxy S3 is only on Android v4.3 the last official.

I do have an Galaxy S5 boxed up somewhere i will have to give it a go.

The pop-up asks every time even though i tick the box on my S3

John.

[eecs]

The pop-up asks every time even though i tick the box on my S3

Did the same for me on my S5.

[Tinderbox (UK)]

I tried my S5 and it works fine apart from the pop-up and it works even if the app is still in memory unlike my S3 which is only works now and then if the app is in memory but everytime if it`s ended and then restarted.

The seek software versions are the same version on my S3 and S5.

John.

[IwuzBornanerd]

The pop-up asks every time even though i tick the box on my S3

Did the same for me on my S5.

Same here on Androidx86.