Hi, i am facing some issues with the enrichment. i downloaded the raw data and dropped them in pupil player, but i am not getting any video or audio. not gaze data nothing. how do i rectify this?

Hi! I have a problem with my Invisible device. It is not recording the scene video. I tried the same a few days ago - it didn't work and then I saw the following notification (see the attached picture)

shoutAttached:sensor is not available:UVCMicSensor

However, when I tried it today - the problem remains but I didn't manage to reproduce the error notification.

Could you please help me out?

Yes, I do. One sec..

It didn't work.

I cannot see the preview when the glasses are connected. I can still press "RECORD", but I cannot see the video later with the message ."no scene video"

Sure.

this is what I see.

When I first opened the app, I also got this notification

OK, thanks!

@user-d407c1 I am using window 11 & chrome

@user-d407c1 oh ok thanks. is there a possible way that i can download induvial users heat map from the enrichment, togeather?

@user-d407c1 Hello again, i have done the above mentioned no doubt its amazing. but i wish to know if there is a way where i can download all individual heat maps of all users in one go, rather than searching them one by one and downloading them. Oh and BTW i no its late but still, happy new year to you and your team 😀

@user-d407c1 also i have recently found that some users enrichment data has not been generated. how do i rectify this?

@marc Noted the points you mentioned and they haven't been the case for the issues. the upload was done on the day of the event and enhancement has been conducted the day after. So as per solution, i would really appreciate if a member of your team can help me out on the work space. i would be able to clear some more smaller doubts as well.

Hi, I'm trying to use the pupil-labs-realtime-api on Jetson but facing dependency errors (cchardet at the moment). Is there a thread for installing Pupil API dependencies on ARM setups?

Hi, when I wear pupil invisible glasses, it can not detect worn. It always prints false. Why this happen? Is it the hardware problem?

Hi happy new year

I have a quick question what is the range of gaze coordinates in pupil labs data?

That would be as follows for the different products: Neon 132°x81° Invisible 82°x82° Core 99°x53°

No, I meant the gaze data that is sent by Pupil invisible API

what are ranges of x and y?

Ah, okay! Those are gaze coordinates in the scene video in pixels. The resolution of the scene image is 1088 x 1080 pixels.

To a small extend Pupil Invisible can estimate gaze points outside of the image boundaries, so the possible values in practice can go slightly out of bounds of that.

Thanks a lot. it was very helpful.

Hello folks, Quick one: Has anyone had the vocals of their invisible recordings just go off? Currently experiencing no sound in some of my workspaces. Is this recoverable? What am I missing please? Cheers.

P.S. recent recordings have been done by a colleague - I am reviewing remotely. atm. Older vids, done by me seems to be having the same issue as well... 🤔

Hey guys, I'm trying to run the reference image mapper enrichment in the cloud project editor, this progress bar has been stuck like this since I uploaded it half an hour ago - are you guys getting similar processing times?

Update - with a 2:14 reference "scan" and a 14, 16, & 17 second test run, it took the cloud 46 minutes to give me a heatmap. Interested to hear if you guys are getting similar results

Hi @user-cd03b7! Just wanted to add a few points, in addition to @user-d407c1's notes about checking progress and refreshing browsers. The reference image mapper is quite computationally intensive, which is why we employ Cloud computing to run it! There are actually two steps to the process, 1. model building and 2. mapping. The model building part takes the longest, but only needs to be performed once per scanning recording + reference image combination. When you're making scanning recordings of larger features, e.g. in the built environment (https://docs.pupil-labs.com/invisible/enrichments/reference-image-mapper/#_4-an-entire-building), it can take some time to complete, and what you describe sounds reasonable in my experience.

Hi, was wondering where to provide feedback on the UX of the use of the Reference image mapper (beta) enrichment. I'll just put it here for now. Reference image mapper requires to have the scanning recording in the project to be able to point the software which recording tot use for Scanning Recording. The problem is that the recording (although no eyetracking data availlable, hence not wearing teh glasses while recording the Scanning recording) seems to be taken into the analysys for generating the heatmap. When deleting the recording the heatmap is rebuild to a more correct one. Imho the scanning recording should be handled (made visible) more on the background, just like the reference image is. So only visible in the settings of the enrichment and not per se as an item in the project. Another thing is, that it is not clear howto analyse a later added recording. I can add it to the project, but it doesn't get analysed automatically within the enrichment I suppose. But when entering the enrichment tab, there is no way of pushing the 'Play' button again to analyse another recording. Might it be the consequence of deleting the Scanning recording? Have not tried to leave the scanning recording in the project and then adding another recording to be analysed. How does/should this work? Also, it is not possible to hit the 'Play'button in the enrichment, when using a begin and end event defining a reletively short scene (eg. 3 seconds). The button to analyse is then greyed out. Could it be that the selected scene is also used on the Scanning recording, which is then to short to build a model? Update: Forget the last comment. I was using event definitions that weren't set in the recording. I used events from within recordings with different type of activity. Should't scenes be somehow restricted to be used within a project or something like that? Update2: Analysys is only possible if new added recording also has same events as used in enrichment (logically)

Hi @user-6a1b78 !

Thanks for your detailed feedback! We use canny.io for feedback (you can find the link on the docs, the cloud, or directly here at feedback ). I've logged your request for you here (https://pupil-labs.canny.io/pupil-cloud/p/rim-improve-ux-and-scanning-recording-handling), so you don't have to. But you can upvote it!

We are working towards improve the usability of the Cloud, but in the meantime... - Correct Heatmap: You should not remove the scanning recording, but rather when clicking on the enrichment's heatmap, use the filter box (see screenshot below). Only the recordings listed below are used for the aggregated heatmap.

• To add new recordings, simply add them onto the project, the enrichment progress bar will show a new empty section and it will allow you to click the play button to run it on the new recordings. This would not work if the events delimiting the sections on which to run the enrichment do not have the same name.

Hey guys, couple quick reference image mapper questions -

1. For my reference scan, I have a couple people moving in the distance of the recording - will this impact heatmap/scan accuracy? I eventually move closer to scan the areas where people were moving (obviously scanning now that they're gone), but is this something I should be worried about?

2. How does occlusion mitigation work if there are other people in the tester recordings? Testers seem to occasionally look at other people's backs as they navigate through the environment that was scanned without people.

3. My current heatmapping effort includes seven tester recordings, one very thorough area scan, and six reference images - how should I go about event creation? The explainer on the pupil site doesn't cover this very well, and the Alpha Lab RIM explainer doesn't cover this either - do I need to create events for each of the six reference images in the scan recording, in each of the tester recordings, or both?

4. Can heatmapping for one reference image overlap with other reference images? My testers are walking along a path and don't always focus on one reference image at a time, so I'm worried that if I isolate reference image events to a specific start and stop time it'll inhibit heatmapping for other reference images - ie, a tester is walking past a sign, so I event stamp that time period as "sign", but during that period when they SHOULD be looking at the sign they're looking at an escalator, which is in a totally different event stamp.

Hey @user-cd03b7! Responses below: 1. This is often unavoidable in the wild for sure. So long as you captured enough data when the people had gone, it shouldn't matter in my experience!

1. Good question! This is definitely a factor that might need manual intervention. If you think it's going to affect your results significantly that is. Unfortunately, there isn't a super easy solution for this just yet. Although we are working on finding one. Check out these manual workarounds: https://docs.pupil-labs.com/invisible/enrichments/reference-image-mapper/#occlusions

2. You'll need to create a separate RIM enrichment for each reference image. Then there are actually two approaches: 3.1. You could create a start + end event that corresponds to when a participant was within the respective area. You'd just create, e.g. start-sign and end-sign. You'd use these same events for all your participants, as presumably they all walked the same route. This focuses the enrichment on a smaller part of each recording (i.e. when each participant walked past the sign), thus saving some computation time. 3.2. You could also just forget the events and run each enrichment on the whole recording. It will just take longer to run.

In Alpha Lab (https://docs.pupil-labs.com/alpha-lab/multiple-rim/#map-and-visualize-gaze-onto-multiple-reference-images-taken-from-the-same-environment), we took the first approach.

1. Each enrichment will only look for its respective reference image. Overlapping in this sense should not be a problem.

Or do scans just get imported into the project and selected in the enrichment section as a raw recording?

Sounds great, thank you!

Last thing - if I have multiple scans of the relatively same area (with ample overlap, the three minute scan length stipulation restricted me from scanning the whole environment), is there a way I can select multiple scans for one reference image in the enrichment? Right now I only see a way to select one scan video.

Quick question for you guys, I'm curious if anyone's running into a similar issue -

I made two RIM enrichments, each using the same two testers, same environment scan, and same reference image for heatmapping. The only difference between the enrichments is that one is using a start+stop events for when the tester was in the area of interest, while the other uses the raw tester recording (recording.begin and recording.end). On paper, I should've gotten the same heatmap, but in reality I got drastically different heatmaps (below).

The image above is the heatmap I got using event enrichments for when the tester was in the area of interest / in proximity to the reference image zone (signage.begin and signage.end, each roughly 15 seconds apart from each other), the heatmap below is using the "raw" tester recordings (pulling from recording.begin and recording.end, each roughly 2 minutes apart from each other)

Is this normal? Is using recording.begin and recording.end for RIM enrichments supposed to give relatively skewed results? Not sure if I made a mistake somewhere or is someone else is running into similar issues.

It may be worth noting I reduced the "scale" factor in the heatmapping settings bar from 5 to 3 for the above images.

I went back and looked at the scene/reference image view, the mapping seems to be a bit off... it's like this for most of the recording, the red fixation visualizer on the right doesn't align with where the tester is looking on the left.

Hey @user-cd03b7! Firstly, thank you for sharing a detailed overview of how you are using the Reference Image Mapper, including screenshots. It is fascinating and pretty awesome to see it being leveraged in this real-world scenario and your feedback is very helpful. Now, let’s address some of your questions! 1. It's only possible to use one scanning recording per reference image at the moment 2. Judging by the screenshot you've shared, we think that something may have gone awry 🤔. Would you be able to ping me the enrichment id via DM and give the Cloud team permission to debug it?

Hi there, Is there a limit to the amount of data we can upload to Pupil Cloud? We're trying to upload a long recording (3GB) to Pupil Cload and it fails and the internet speed is good and stable.

Does Pupil Invisible LSL Relay work for any pupil core device or only pupil invisible?

Only Pupil Invisible (and in the future for Neon). For Pupil Core, check out https://github.com/labstreaminglayer/App-PupilLabs/tree/master/pupil_capture

Second question. Does the Pupil Core pupil_capture work to stream real-time data?

Hey, how have you guys been moving recordings from one workspace to another? I have a coworker who wants to check out some of my recordings but they're on a separate account, and I don't see anything that allows me to upload a recording

Thanks for your reply! No it doesn't even start uploading. Could there be a problem with the account? Another device using another workspace doesn't have a problem uploading. We have uploaded recordings on this workspace before but it's not working anymore.

Hello, is this the appropriate place to ask questions about the Pupil Labs Cloud?

ok thank you, I currently have 2 recordings on my pupil cloud that are listed as 99% uploaded, but have remained at this stage for a number of days now. Is there a way I can resolve this?

thank you

Hi not sure if this is the right channel to post in - however... I am trying out AOIs in the Reference Image using the instructions you provide here (https://docs.pupil-labs.com/alpha-lab/gaze-metrics-in-aois/). It worked perfectly well using your workspace data/fixations however... when trying to use our data the fixations.csv that gets exported is missing the column that says "fixation detected in reference image". When I export your trial data directly from the workspace cloud, it is also missing this column... however the file you uploaded has this column. Just wondering if there is a special option/setting that needs to be selected so that the csv file that is exported directly from the cloud includes this column of data. I have attached fixationsExample1 (with the column - what you use as an example csv for the trial of paintings) and fixationsExample2 (without the column - what is created when you download the data directly from the cloud)

I will check the recordings on our end tomorrow at work - but for now the ID of the workspace video is b561aeee-7cc8-4a9f-9840-930db6677a66 and it is labelled Jack_Standing_Upstairs

We are basically just trying to upload the fixations.csv to the python code for the AOIs but without that column of data being exported within the csv it won't work

Oh fantastic - I knew it was something I was doing incorrectly... thank you so much that was really helpful 😄 P.S. this feature is AMAZING! Our lab is very impressed 🙂

Hey guys, quick RIM question - when I run two identical enrichments in the same project enrichment editor space (same two testers, same start and stop events, and same reference image and scan recording, just different enrichment names), I get two slightly different heatmaps -

Again, fairly negligible differences, but on paper I should be getting identical heatmaps here, right? Did I do something wrong?

Follow up question/issue - I ran the following two RIM enrichments with the same three tester recordings and the same environment scan, but each with a different reference image. These reference images are two halfs the same sign, with one RIM enrichment for each half because I wasn't able to get an angle on the narrow platform to get the whole sign. (note, scale is "5" and colormap is "turbo")

However, when I (finally got smart enough to) stitch the two reference images together to create a reference image of both halfs and run the same RIM enrichment (with the same three tester recordings and environment scan, just with this new "combined" reference image), I get the following output - also with a scale of "5" and Turbo colormap

Considering I'm working with the same test data and environment scan, why is there such a sharp disparity between the separated scans and the combined scan? There's virtually zero activity around the "3rd Ave & Pine" area of the combined RIM, but in the separated RIM there's tons of plotted activity over that section.

And in the uppermost image from the separated RIM, there's heavy activity on the right side of the image over "Westlake", which happens to contain overlapping imagery in the second half of the separated RIM. But in the second half of the RIM, the overlapping imagery doesn't contain the same plotted activity over the text "Westlake". Am I doing something wrong here? Has anyone encountered a similar issue?

Thanks Marc, sounds great! Just sent you a message regarding the first bit, and your explanation on image stitching makes complete sense as well. Thanks for the pointers on this one, I'm gonna poke around and see what I can do!