search

dcnieho / Titta

Matlab and PsychToolbox interface to Tobii eye trackers using Tobii Pro SDK
Other
41 stars 15 forks source link

[Feature Request] Selectively recollect calibration points #17

Closed iandol closed 3 years ago

iandol commented 4 years ago

Babies or animals are not easy to calibrate, and often lose attention momentarily or randomly look too briefly at a target. Currently, we can "redo" a point if we are still in that point with backspace. However it isn't always easy to trigger this at the right time, and often you end up with a mix of good points and bad ones, which causes a calibration to fail. It would be really nice to be able to recollect points selectively. Tobii Studio allowed for this in its infant calibration mode (however Pro Lab or ETM's calibration mode does not, among other things). It would be hugely beneficial to able to select calibration points. Eyelink allows this and uses a fairly simple strategy:

  1. Go into a manual calibration mode.
  2. Press 1-9 to trigger display of that point.
  3. Space to trigger collect data, but after collection responsibility to press the number for next calibration point.

In addition some way to turn off display of a calibration point and go into a "pause", without totally escaping the calibration run.

I had a quick look at DoCalPointDisplay() which is where I assume the logic for this would have to go.

It would be great on an operator screen that the status of calibration points that were collected could be highlighted, or a text line with this info drawn. I realise this information may not be available however, calibrationCollectData() doesn't seem to return any status (possibly because the C SDK returns a status that only means no error?). EDIT: OK I see the status is coming from calibrationRetrieveResult()...

iandol commented 4 years ago

Rough flow diagram of the fully manual calibration loop flow

dcnieho commented 4 years ago

@iandol, interesting request! Would be willing to look into that. Lets see (read the whole thing first, this is partially a brain dump):

Ok, to clarify:

Points to think about:

Design that i am thinking about:

Note that this way you can define way more points for calibration and validation in the settings than you intend to use, and then select during calibration which you'll actually use. This allows you to e.g. collect for a smaller grid if points in the corners didn't work.

Describing the practical issues, practical tricks and interface design for such a manual calibration procedure may well make for an interesting paper. There is also a trade off between complexity of interface and training of assistants. I can think of a colleague working with infants who may well be interested in this as well. Email me offline if you want to pursue that further (if you see a need/use for such things in your field), but lets keep discussion of the interface in the thread here.

iandol commented 4 years ago

Thanks for a thorough reply! ;-)

only makes sense on two-screen setup, right?

Actually, with the Eyelink, although it always has an operator mode (dedicated PC with realtime OS runs the show), there is no real interface for manual calibration, you select points without almost any feedback (forgetting if top-left is 1 or 5 etc), and it is a klunky affair. So even having a text log in the command window of which points were selected to test and which completed would be a hell of a lot better than the Eyelink. I suspect it would be easier for you to do. However online visual feedback really improves the experience tremendously, so two-screens would definitely be preferred.

Would need to somehow mark each point if not yet collected or unsuccessful collection

Yes, an important part of getting manual calibration to work, although as you said, it may be that even 3 points in a 5 point calibration would give a succesful calibration. I'm normally used to thinking that if it is 5 point calibration, all 5 must be completed (as is the case for eyelink).

when no point selected, calibration display for participant should not show anything

Yes, sometimes we need to take a small time out and give a bit of apple or something, so stuff on the screen would be confusing.

Note that i have never seen Tobii Studio, so do not know what that looked like or what their mechanism was

As we are having problems with the firmware with our Spectrum Pros (why we can't calibrate at the moment so it turns out), we just started testing the TX300 and Tobii studio today, I'll be able to give you some more feedback on how infant mode is controlled tomorrow.

Related: want the option to recalibrate specific points after validation?

Without us knowing which points are causing a poor validation I suppose it is hard to know which points, but I imagine there must be a spatial relationship with validation errors and their corresponding calibration points, so this may be a nice option to have.

There is no option to go back to automatic (we can use screen real estate better).

Sounds reasonable to me.

bet you don't want to show it to them

Correct!

After each collected data point, we call the compute calibration function

I hadn't thought of that but that is cool if it is possible and doesn't interrupt calibration

You can enter a delete mode

It could be like shift-click or shift-number for discard+recollect.

Need to think about how to enable menu selecting among multiple calibrations

Because you can switch iteratively between cal and val quite easily, not sure how we define a previous calibration?

I'm happy to help where needed on this, and I think it will really benefit those of us working with non-verbal subjects.

iandol commented 4 years ago

FYI from the Tobii Studio manual (sorry for poor formatting from PDF):

Infant calibration is used when conducting studies on infants.Small animations(in AVI format)are displayed where the dots nor-mally appear.This works as an attention grabber suitable for small children.It can also be used when it is difficult for someone to follow the calibration instructions. TobiiStudio can alsodisplay different AVIsto attract the test participant’s attention if they begin to respondpoorlyor growtiredof the calibrationAVI.Theseitemsare calledattentionanimationAVIs.No calibration datais collected duringthe displayoftheseAVIs.Codecsthatcanbe used to make the AVI files are MicrosoftVideo1, Techsmith1or Xvid

Infant calibratio nis an interactive calibration mode where an operato uses the keyboard to contro lthe calibration procedure.To show the calibration animation selected in the GlobalSettingsdialogbox, press the RIGHTARROW key during the calibra-tion. When the participant is watchingthe animation, press the RIGHTARROW key againto start collecting calibration data. Itis assumedthatthe participantwill focuson the center of the calibration animation. When enough datapointsare collected,thecalibrationwill move on to the nextcalibrationpoint.The operator can again press the RIGHTARROW key to start collecting data.Thisis repeated untilall calibration points have been displayed.The following keys are used to control the Infant calibration procedure: •RIGHTARROW: Displays the calibrationAVI at the nextcalibrationpoint.If the calibration AVI is alreadyvisible,a cali-brationis madeat thatpointbeforeproceedingto the next. •LEFTARROW: Pause. Hides the calibration AVI or the full-screen attention-grabbing AVI •UP ARROW :Displays the attention grabbing AV

The UI for selecting points to recalibrate is simple:

Untitled

Click the points that you want to recalibrate and only those are tested. During calibration it provides no feedback on where the raw eye trace is or anything else (it only uses a single screen).

dcnieho commented 4 years ago

Thanks for the discussion and info about Tobii Studio.

I think a combination of what Studio does and what the eyelink does, with a bit of own special sauce, will make for a very nice interface. There would be visual feedback all the time (on operator screen) about what is happening, and recalibration is as simple as selecting an already collected point.

Yeah, the call to compute the calibrate is really cheap on most of the eye trackers i have tested with (exceptions that i remember are the 4C and X2-60, but so be it for users of those, I consider it pretty niche that those systems would be used with difficult populations--though i don't have stats). Its necessary to call the calibration function to get that line display like we have on the validation result screen and also in the Studio screenshot.

Regarding number of calibration points, i guess that for this advanced use case, it would be nice to offer the possibility to skip a point that can't be calibrated and still continue. Practice will show how robust the eye trackers are to that.

I'm not sure how to define a "calibration" indeed. Either any state that precedes some collection of validation data, or perhaps have an explicit "save" button or so. Probably the latter is better, more flexibility. Downside might be that some think it must be pressed, but there would be no cost to that except the extra operator action.

Regarding delete, I think it should be distinct from recollect. So two things can happen on an already collected point:

  1. recollect: normal click
  2. delete: indeed like shift click or so. Its perhaps a bit niche, but you may find that a specific point, even though some data was collected, is giving you trouble, so you want to just kick it out again. Not sure that ever happens, but why not offer the flexibility, as its super cheap to do.

Thanks for the input! I think perhaps the next step is for me to play with this, in a separate branch, so we have some mock up or so that you can test as well and see what you think.

iandol commented 4 years ago

Regarding number of calibration points, i guess that for this advanced use case, it would be nice to offer the possibility to skip a point that can't be calibrated and still continue. Practice will show how robust the eye trackers are to that ... even though some data was collected, is giving you trouble, so you want to just kick it out again...

Actually, single calibration points not calibrating is a pretty common experience for me at least, even sometimes for humans wearing glasses.

I think having an explicit "save" allows an operator to "tag" a state, it would enable them to have more control, and you could call it something like "snapshot" to identify it as optional?

dcnieho commented 4 years ago

Good idea to call it a snapshot! Ok, this has taken enough form now that I'll try to implement a first attempt.

@iandol, please do update to latest version of Titta now as well. I found that there was some logic error when you have multiple calibrations. In some situations, the validation result screen may not show the one currently selected. Also, in some cases Titta.calibrate() does not log the applied calibration in a message, even though one was applied. These are now fixed. Titta also gained a new option on the setup screen (with the head), press 'c' to change from which eye you are recording (if your eye tracker supports that). Its a by-default hidden button on that interface, but the accelerator key always works. Implemented this for another type of difficult subject, e.g. with squint.

iandol commented 4 years ago

Ha, indeed one of our animals has strabismic squint, so this is potentially useful when we get our Spectrum Pro back working (our interim TX300 doesn't support monocular calibration right?)... Will have a test, thanks Dee.

dcnieho commented 4 years ago

Hey! Its by far not done yet, but give the dev branch a try if you are curious. Don't use the snapshot menu button, it'll crash. Deleting point data is also not possible. Also, i haven't tested on Linux, but don't see why it wouldn't work :P Run demos/readmeManualCalibration.m

What you can do:

I don't have any specific things i'm looking for feedback on yet, but do tell me of course about thoughts that come up. Ah yeah: a known bug is that point collection status is not reset when changing to monocular calibration mode, or back. Just not implemented yet (shit there are many interacting parts to think about here!)

Oh, i see i missed a question. Indeed i think the TX300 doesn't do monocular calibration. But easy enough to check, if it does, the eye selection menu would be available in Titta, if it doesn't that menu isn't available.

iandol commented 4 years ago

Hi Dee, I had a quick play just now, but even though I changed the presenter/operator screens (in linux/mac these are 0 and 1, not 1 and 2 as on Windows), somewhere in the drawing code, most of the operator display UI ended up in the presentation display anyway. I can step debug my way through but don't have the time to test more until around Wednesday...

dcnieho commented 4 years ago

Hi Ian,

Thanks for giving it a try! Are you sure you did not swap around the participant and operator screens? I just looked through the code, and for instance the buttons should show up on the operator screen, i don't see how they could end up somewhere else. I would really appreciate it if you could look at why things end up on the wrong screen for you on Linux, when you have time.

On Mon, May 11, 2020 at 8:47 AM Ian Max Andolina notifications@github.com wrote:

Hi Dee, I had a quick play just now, but even though I changed the presenter/operator screens (in linux/mac these are 0 and 1, not 1 and 2 as on Windows), somewhere in the drawing code, most of the operator display UI ended up in the presentation display anyway. I can step debug my way through but don't have the time to test more until around Wednesday...

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-626505920, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGPOK4NNRDI7HCRAGK3RQ6NQJANCNFSM4LDOQFLA .

iandol commented 4 years ago

Yes indeed, operator error ;-) I just defaulted to operator screen is going to be the first display and presentation the highest value returned by Screen('Screens'), but you do it the other way round, and I didn't check so got the numbers the wrong way round.

This is great!!! You've got most of this working the way I envisaged already, and being able to see as we collect points what the overall status is and recollect points is great. Some questions:

1) If a point is green but doesn't have the error lines, does that mean we tried to collect that point but no calibration data was valid? If so perhaps have an intermediate colour to more quickly identify these point. 2) Some points I'd collected went back to grey after collecting other points, perhaps they were points that were collected but didn't have valid data, they reset back to grey? 3) I do really like how you've managed to combine the two interfaces on the operator screen, the floating resizeable setup panel is super cool!!! Lowish priority, but we normally need to see the head position setup on the presenter as well as operator screen (this is currently an option in Titta master, usually if we have one operator and we have separate rooms for subject/operator they have to move our chair around and seeing it on the screen in the room where the subject is helps). I realise one problem is that the setup panel could obstruct calibration points if the operator forgets to toggle it off, and this may cause confusion for the operator. I'm not sure the best way round this (remind operator to toggle setup-panel off when collecting data points somehow, or always show setup-panel on operator screen and toggle it on/off only on presenter screen).

Let me know what you would like me to focus on if I can help with targeted testing. But this interface is already way better than anything I've used with the Eyelink or Tobii Pro Studio/Lab 👍 😀

dcnieho commented 4 years ago

Hi Ian,

Great it works and that it's making you happy already! Let me answer your points in alternative order:

  1. This surprised me at first two, but some orders of calling lead to different results: Doing first only the top left point leads to calibration data collected (point green), but calibration failed (see top left of screen). Then collecting the point in the middle leads to calibration success, but that computation removes the top left point apparently. I can see that by the calibration feedback i get from the Tobii, so i made that point grey again. Doing first middle, then top left, both are used for the calibration. Doing top-left and middle in one sequence (press them quickly together, so second gets queued and shown directly after first, also leads to both points being used for calibration. Nothing i can do about that i guess, except communicate it the way i do now (let me know if you think of a better way!) and operators will have to learn these dependencies i guess. Note that I don't know yet if the spectrum always throws out data once calibration fails, or only in the case described above. Its hard to get it to fail after the center has been collected :p
  2. Is this the same point as the previous question? Or perhaps not: this may happen if you collect data for a point (e.g. top-left only) and calibration fails. Then there is data for the point, but no lines to show as they are generated by the calibration computation. I assume data for a point is collected (if the point collection function succeeds, else they turn red) until i hear otherwise from the calibration function (see above). If you are using a spectrum again, can you describe a specific sequence where you see this? Then i can see if i can reproduce if it is not expected (NB, for as far as i know what to expect :P).
  3. If you hold down shift while pressing the head button or keyboard key, you should get the head display on both screens. It disappears on the subject screen when a calibration point is shown and reappears after (not sure why, i didn't implement that, might only be with the animated drawer, need to look further into that). Does that do what you need? If operator wants to see the screen but no longer to participant, toggle it off and back on without holding shift down.

On Tue, May 12, 2020 at 2:08 AM Ian Max Andolina notifications@github.com wrote:

Yes indeed, operator error ;-) I just defaulted to operator screen is going to be the first display and presentation the highest value returned by Screen('Screens'), but you do it the other way round, and I didn't check so got the numbers the wrong way round.

This is great!!! You've got most of this working the way I envisaged already, and being able to see as we collect points what the overall status is and recollect points is great. Some questions:

  1. If a point is green but doesn't have the error lines, does that mean we tried to collect that point but no calibration data was valid? If so perhaps have an intermediate colour to more quickly identify these point.
  2. Some points I'd collected went back to grey after collecting other points, perhaps they were points that were collected but didn't have valid data, they reset back to grey?
  3. I do really like how you've managed to combine the two interfaces on the operator screen, the floating resizeable setup panel is super cool!!! Lowish priority, but we normally need to see the head position setup on the presenter as well as operator screen (this is currently an option in Titta master, usually if we have one operator and we have separate rooms for subject/operator they have to move our chair around and seeing it on the screen in the room where the subject is helps). I realise one problem is that the setup panel could obstruct calibration points if the operator forgets to toggle it off, and this may cause confusion for the operator. I'm not sure the best way round this (remind operator to toggle setup-panel off when collecting data points somehow, or always show setup-panel on operator screen and toggle it on/off only on presenter screen).

Let me know what you would like me to focus on if I can help with targeted testing. But this interface is already way better than anything I've used with the Eyelink or Tobii Pro Studio/Lab 👍 😀

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-627031004, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGMCSDQHTY7YKXRGZNDRRCHQBANCNFSM4LDOQFLA .

iandol commented 4 years ago

Yes question 1 and 2 were basically the same question, and great that you you can use shift to toggle, that sounds fine (can't test today)...

dcnieho commented 4 years ago

hey man, its basically done. Check out the dev branch. Only the functionality to load previous calibrations when calling the calibrate function is not implemented. Snapshot work. and it turns out they're really snapshots: when you load one, you are back in the state when you saved it, and you can add points to it. So you can use them to store something you are happy with before you continue and try to improve, and if your attempt to improve backfires, nothing is lost. so shift-click on points to discard their data three things to note:

  1. when a calibration fails, the previous one stays active. Hence the status message in top left now contains info about which points a currently active calibration is based on. So if you see "calibration failed" this means the last calibration attempt failed, not that you are not calibrated at all. you are not calibrated at all only when there are no points listed
  2. if you calibrate piecemeal, tobii recommends to start with either a center point, or two points at opposite sides of the screen. I have also noted something about the following order (for my eyes). if i first do point 3 (middle), let it calibrate, and then queue up points 1 and 5 (top-left, bottom-right) immediately, sometimes this yields a successful second cal, sometimes not. if not, recollecting 3 usually fixes it. Anyway, it seems there is a lot of getting to know your machine here before you know how to best approach such a flexible calibration for your tracker
  3. snapshots only store a calibration state. You can later add validation for that calibration, and the snapshot menu will be updated to reflect that. Note also that if the currently active calibration corresponds to an item in the snapshot menu, that item is highlighted. If not, no item is highlighted.

I've reasonably tested this on Windows, and expect it to work on Linux too. I hope you can seriously put it through its paces, see if you can get it to crash or act unexpectedly. Do send me the callstacks for any crashes and ideally also how to reproduce them (you can send these by mail, they may be long). I'm of course also very interested to hear about how this actually works for you.

On Wed, May 13, 2020 at 9:42 AM Ian Max Andolina notifications@github.com wrote:

Yes question 1 and 2 were basically the same question, and great that you you can use shift to toggle, that sounds fine (can't test today)...

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-627809403, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGLUKCUZWGAEV25UMMLRRJFONANCNFSM4LDOQFLA .

dcnieho commented 4 years ago

If you at some point test with a monkey, would you be able to make a screen recording of the process? I would be very interested to see how you actually use it.

On Sat, May 23, 2020 at 10:48 PM Diederick C. Niehorster dcnieho@gmail.com wrote:

hey man, its basically done. Check out the dev branch. Only the functionality to load previous calibrations when calling the calibrate function is not implemented. Snapshot work. and it turns out they're really snapshots: when you load one, you are back in the state when you saved it, and you can add points to it. So you can use them to store something you are happy with before you continue and try to improve, and if your attempt to improve backfires, nothing is lost. so shift-click on points to discard their data three things to note:

  1. when a calibration fails, the previous one stays active. Hence the status message in top left now contains info about which points a currently active calibration is based on. So if you see "calibration failed" this means the last calibration attempt failed, not that you are not calibrated at all. you are not calibrated at all only when there are no points listed
  2. if you calibrate piecemeal, tobii recommends to start with either a center point, or two points at opposite sides of the screen. I have also noted something about the following order (for my eyes). if i first do point 3 (middle), let it calibrate, and then queue up points 1 and 5 (top-left, bottom-right) immediately, sometimes this yields a successful second cal, sometimes not. if not, recollecting 3 usually fixes it. Anyway, it seems there is a lot of getting to know your machine here before you know how to best approach such a flexible calibration for your tracker
  3. snapshots only store a calibration state. You can later add validation for that calibration, and the snapshot menu will be updated to reflect that. Note also that if the currently active calibration corresponds to an item in the snapshot menu, that item is highlighted. If not, no item is highlighted.

I've reasonably tested this on Windows, and expect it to work on Linux too. I hope you can seriously put it through its paces, see if you can get it to crash or act unexpectedly. Do send me the callstacks for any crashes and ideally also how to reproduce them (you can send these by mail, they may be long). I'm of course also very interested to hear about how this actually works for you.

On Wed, May 13, 2020 at 9:42 AM Ian Max Andolina notifications@github.com wrote:

Yes question 1 and 2 were basically the same question, and great that you you can use shift to toggle, that sounds fine (can't test today)...

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-627809403, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGLUKCUZWGAEV25UMMLRRJFONANCNFSM4LDOQFLA .

iandol commented 4 years ago

Yes, I'm happy to show you it in action with some subjects via mail.

I get a crash when using the snapshot menu sometimes (possibly pressing twice), callstack is:

Unrecognized function or variable 'iValid'.

Error in Titta/doManualCalib (line 3920)
                        nElem           = length(iValid);

Error in Titta/calibrateManual (line 1065)
            out = obj.doManualCalib(wpnt,out,kCal,currentSelection);

Error in readmeManualCalibration (line 124)
    tobii.calVal{1} = EThndl.calibrateManual([wpntP wpntO]);

I'll work round it and continue testing.

dcnieho commented 4 years ago

Thanks! Did you check out the latest dev branch HEAD? That code doesn't appear on that line anymore, it looks like its about a week old or so.

On Tue, May 26, 2020 at 8:09 AM Ian Max Andolina notifications@github.com wrote:

Yes, I'm happy to show you it in action with some subjects via mail.

I get a crash when using the snapshot menu sometimes, callstack is:

Unrecognized function or variable 'iValid'.

Error in Titta/doManualCalib (line 3920) nElem = length(iValid);

Error in Titta/calibrateManual (line 1065) out = obj.doManualCalib(wpnt,out,kCal,currentSelection);

Error in readmeManualCalibration (line 124) tobii.calVal{1} = EThndl.calibrateManual([wpntP wpntO]);

I'll work round it and continue testing.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-633825992, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGP5KT6CT76QNH4LWL3RTNMI3ANCNFSM4LDOQFLA .

iandol commented 4 years ago

I'm on this commit according to Git:

Titta on  dev 
[I] ➜ git log -1
commit 347b0258e1d11a4e00ce136665e32cb1fe66f42c (HEAD -> dev, upstream/dev)
Author: Diederick Niehorster <dcnieho@gmail.com>
Date:   Mon May 25 13:18:39 2020 +0200

    snapshots didn't work when there were discarded points

But something seems to have gone wrong with my local repo, so I recloned and it seems to be back in sync. Don't see that error now and indeed snapshot menu is working great!!! Workflow is really great. There do indeed seem to be some strange order of collection issue, I'll try and see if I see a similar pattern to you. We'll probably give this a quick test with subjects tomorrow and I'll give you some more feedback then.

It's really nice, thank you for all your hard work on making this interface a reality!

dcnieho commented 4 years ago

good to hear and thanks! Yeah, we're just working with what the eye tracker does, the interface does its best to report what the actual state is, no matter how you got there. I guess it'll be important for your operators to have experience with the system and know what they should and shouldn't do to not lose points (since points are precious). I'm of course happy to hear about the things you learn and see if i can reproduce them! And thanks for the thanks! It was quite fun and hope it'll be useful for you and more labs

On Tue, May 26, 2020 at 12:44 PM Ian Max Andolina notifications@github.com wrote:

I'm on this commit according to Git:

Titta on dev

[I] ➜ git log -1

commit 347b0258e1d11a4e00ce136665e32cb1fe66f42c (HEAD -> dev, upstream/dev)

Author: Diederick Niehorster dcnieho@gmail.com

Date: Mon May 25 13:18:39 2020 +0200

snapshots didn't work when there were discarded points

But something seems to have gone wrong with my local repo, so I recloned and it seems to be back in sync. Don't see that error now and indeed snapshot menu is working great!!! Workflow is really great. There do indeed seem to be some strange order of collection issue, I'll try and see if I see a similar pattern to you. We'll probably give this a quick test with subjects tomorrow and I'll give you some more feedback then.

It's really nice, thank you for all your hard work on making this interface a reality!

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-633949577, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGJ2PAE6BWU5HSHNE7DRTOMR7ANCNFSM4LDOQFLA .

iandol commented 4 years ago

Here are a few observations:

1) [shift]+[h] doesn't seem to work anymore to show head on both displays, although [shift]+click button does. 2) calibration.selectedCal used to return a single number, but now it seems to return something like [1 X], even if there is only one cal in the attempt array:

           type: 'manual'
    selectedCal: [1 5]
     wasSkipped: 0
          flips: [1×4206 double]
        attempt: {[1×1 struct]}
         status: 1

3) Trying to run getValidationQualityMessage(calibration) tells me:

'Reference to non-existent field 'status'.

     Error in Titta>@(x)x.status (line 1826)
                     iVal    = find(cellfun(@(x) x.status, cal.val)==1,1,'last');

     Error in Titta.getValidationQualityMessage (line 1826)
                     iVal    = find(cellfun(@(x) x.status, cal.val)==1,1,'last');

4) Are you planning to add back the ability to pass a previous calibration or is that now not possible for manual mode? 5) It rocks! ;-)

dcnieho commented 4 years ago

Thanks!

  1. thats a regression, will fix
  2. yes, for manual calibration, it returns which attempt, and within that attempt, which cal (note there is a cell-array 'cal' in each attempt
  3. Ah, you can't pass one of these calibrations directly to that function. I agree that that should be possible. will fix
  4. yes, thats the one todo left.
  5. great!

On Tue, May 26, 2020 at 1:20 PM Ian Max Andolina notifications@github.com wrote:

Here are a few observations:

  1. [shift]+[h] doesn't seem to work anymore to show head on both displays, although [shift]+click button does.

  2. calibration.selectedCal used to return a single number, but now it seems to return something like [1 X], even if there is only one cal in the attempt array:

       type: 'manual'

    selectedCal: [1 5]

    wasSkipped: 0

      flips: [1×4206 double]

attempt: {[1×1 struct]}

 status: 1

  3. Trying to run getValidationQualityMessage(calibration) tells me:

'Reference to non-existent field 'status'.

 Error in Titta>@(x)x.status (line 1826)

                 iVal    = find(cellfun(@(x) x.status, cal.val)==1,1,'last');

 Error in Titta.getValidationQualityMessage (line 1826)

                 iVal    = find(cellfun(@(x) x.status, cal.val)==1,1,'last');
  1. Are you planning to add back the ability to pass a previous calibration or is that now not possible for manual mode?
  2. It rocks! ;-)

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-633964253, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGJPJAW7LGJAPK3UYSDRTOQYNANCNFSM4LDOQFLA .

dcnieho commented 4 years ago

hi Ian,

Don't pull any new versions from the dev branch anymore, just continue testing with what you have now (unless the inability to show the head on the participant screen is blocking you--let me know i that is the case). I am pushing some untested things, so the HEAD may be broken.

On Tue, May 26, 2020 at 2:10 PM Diederick C. Niehorster dcnieho@gmail.com wrote:

Thanks!

  1. thats a regression, will fix
  2. yes, for manual calibration, it returns which attempt, and within that attempt, which cal (note there is a cell-array 'cal' in each attempt
  3. Ah, you can't pass one of these calibrations directly to that function. I agree that that should be possible. will fix
  4. yes, thats the one todo left.
  5. great!

On Tue, May 26, 2020 at 1:20 PM Ian Max Andolina notifications@github.com wrote:

Here are a few observations:

  1. [shift]+[h] doesn't seem to work anymore to show head on both displays, although [shift]+click button does.

  2. calibration.selectedCal used to return a single number, but now it seems to return something like [1 X], even if there is only one cal in the attempt array:

       type: 'manual'

    selectedCal: [1 5]

    wasSkipped: 0

      flips: [1×4206 double]

attempt: {[1×1 struct]}

 status: 1

  3. Trying to run getValidationQualityMessage(calibration) tells me:

'Reference to non-existent field 'status'.

 Error in Titta>@(x)x.status (line 1826)

                 iVal    = find(cellfun(@(x) x.status, cal.val)==1,1,'last');

 Error in Titta.getValidationQualityMessage (line 1826)

                 iVal    = find(cellfun(@(x) x.status, cal.val)==1,1,'last');
  1. Are you planning to add back the ability to pass a previous calibration or is that now not possible for manual mode?
  2. It rocks! ;-)

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-633964253, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGJPJAW7LGJAPK3UYSDRTOQYNANCNFSM4LDOQFLA .

iandol commented 4 years ago

Yeah I just saw the changes, I can just detach my HEAD to the last safe commit. By the way, I did test your *untested*: implemented that getValidationQualityMessage works on the output of calibrateManual() but it errors out at the moment (untested as it says on the tin ;-), so I just detach to the previous to that commit.

dcnieho commented 4 years ago

good, that should probably work. The commit before is also unteste, but very low risk. It should bring back holding down shift to show head (and online gaze) on both screens.

On Wed, May 27, 2020 at 3:22 AM Ian Max Andolina notifications@github.com wrote:

Yeah I just saw the changes, I can just detach my HEAD to the last safe commit. By the way, I did test your untested: implemented that getValidationQualityMessage works on the output of calibrateManual() but it errors out at the moment (untested as it says on the tin ;-), so I just detach to the previous to that commit.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-634367628, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGNRL3FANDBMBZBKBVDRTRTMBANCNFSM4LDOQFLA .

iandol commented 4 years ago

I can confirm [shift]+[h] is working again :+1:

dcnieho commented 4 years ago

nice! dev HEAD should work again now. I made some major naming changes in the mex files, so its probably best to do a clean clone. TODO still is loading previous calibrations

iandol commented 4 years ago

OK, getValidationQualityMessage now works. I also had to rebuild the MATLAB path (as you renamed the folders). You maybe can add a small function like this to your Titta root folder:

https://github.com/iandol/opticka/blob/master/addoptickapaths.m

To automatically regenerate the correct path (removing .git and any other path locations specified via pathExceptions that MATLAB does not need). A bit easier for general users.

iandol commented 4 years ago

A few small points on the settings:

dcnieho commented 4 years ago

thanks for the comments on the settings, will check. I think i found mancal.cal a bit redundant, and thus left that out. Symmetry and logical grouping (proper deepnesting) is better here indeed, will change.

The various bgColors are for participant and operator screen, and backgrounds of overlays, such as the head positioning overlay. I'll document better.

Interesting point regarding adding to path. I always simply addpath(genpath()) and be done with it. Matlab doesn't seem to have an issue with .git subfolders on the path. I'll look into providing this function though (thanks for sending me the example!) to provide a clean way of adding Titta to the path for those who want.

On Thu, May 28, 2020 at 3:20 AM Ian Max Andolina notifications@github.com wrote:

A few small points on the settings:

  • Why settings.mancal.paceDuration, rather than settings.mancal.cal.paceDuration as you also have a settings.mancal.val.paceDuration and it is more similar to settings.cal.paceDuration?
  • What is the difference between settings.mancal.bgColor and settings.UI.mancal.bgColor and for other similar settings?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-635034705, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGOREBFO7RGMKGSJ7GTRTW37FANCNFSM4LDOQFLA .

dcnieho commented 4 years ago

ok, i've done the settings related thing you asked about. will look at paths next, but not today

On Sat, May 30, 2020 at 7:04 AM Diederick C. Niehorster dcnieho@gmail.com wrote:

thanks for the comments on the settings, will check. I think i found mancal.cal a bit redundant, and thus left that out. Symmetry and logical grouping (proper deepnesting) is better here indeed, will change.

The various bgColors are for participant and operator screen, and backgrounds of overlays, such as the head positioning overlay. I'll document better.

Interesting point regarding adding to path. I always simply addpath(genpath()) and be done with it. Matlab doesn't seem to have an issue with .git subfolders on the path. I'll look into providing this function though (thanks for sending me the example!) to provide a clean way of adding Titta to the path for those who want.

On Thu, May 28, 2020 at 3:20 AM Ian Max Andolina notifications@github.com wrote:

A few small points on the settings:

  • Why settings.mancal.paceDuration, rather than settings.mancal.cal.paceDuration as you also have a settings.mancal.val.paceDuration and it is more similar to settings.cal.paceDuration?
  • What is the difference between settings.mancal.bgColor and settings.UI.mancal.bgColor and for other similar settings?

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-635034705, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGOREBFO7RGMKGSJ7GTRTW37FANCNFSM4LDOQFLA .

iandol commented 4 years ago

No worries, Here is the path script I use modified for Titta:

mpath           = path;
mpath           = strsplit(mpath, pathsep);
opath           = fileparts(mfilename('fullpath'));
for i = 1:length(mpath)
    if ~isempty(regexpi(mpath{i},opath))
        rmpath(mpath{i}); % remove any old path values
    end
end
opaths          = genpath(opath); 
opaths          = strsplit(opaths,pathsep);
newpaths        = {};
pathExceptions = [filesep '\.git|' filesep 'src'];
for i=1:length(opaths)
    if isempty(regexpi(opaths{i},pathExceptions))
        newpaths{end+1}=opaths{i};
    end
end
newpaths = strjoin(newpaths,pathsep);
addpath(newpaths); savepath;
disp('--->>> Added Titta to the path...')
clear mpath opath opaths newpaths pathExceptions i
dcnieho commented 4 years ago

@iandol, I have now implemented the reloading of previous calibrations, please give it a try. When reloading, snapshots and validations are lost, just the cal is kept.

Could you put all of this through its paces thoroughly? And if you could take a video of the process showing how you'd use it, please do and send it to me.

This is all now on the master branch.

By the way, the 1.8 SDK is out, but the Linux link still refers to the old 1.7 SDK. I'll update when i can.

iandol commented 4 years ago

Hi Dee, great thank you! We're not doing any more experiments till next week, but I will test this a bit on my spectrum (N=1).

I will try and record something for you next week.

Do you know what the 1.8 update does (didn't see a changelog in the downloads)? EDIT: found it here: http://developer.tobiipro.com/tobiiprosdk/whatsnew.html — so hopefully the timing issue is fixed...

iandol commented 3 years ago

I think we can call this closed ;-)

dcnieho commented 3 years ago

haha :) Working well for you or are there things that have come up now that you've been using it for a while? No guarantee i'll do something about it, but always interesting to hear about

On Thu, Apr 22, 2021 at 6:52 AM Ian Max Andolina @.***> wrote:

I think we can call this closed ;-)

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/dcnieho/Titta/issues/17#issuecomment-824536473, or unsubscribe https://github.com/notifications/unsubscribe-auth/AANUOGIMYXVON262IWTJIDTTJ6TQVANCNFSM4LDOQFLA .