Datasets & Projects Webinar (English) | Organising Transkribus Workflows & AI Training Data

Hello everyone. Welcome to today's webinar on our two newest features, datasets and projects. We will walk you through these two topics today. I will start with a small introduction, then we will look into datasets, projects, and we will also have some time for a Q&A afterwards. If you have questions, you can always drop them in the chat and we will try to answer them during the session. We will also share the slides of today's webinar and we will also send the recording of what we are talking about then later as well. So, today with me here walking you through these topics is Zara on the one hand. Zara is one of our customer success managers, customer success lead. Um [snorts] together with her team, she is there when it comes to supporting customers, working together with customers on different projects. And um then also Helena is here, our marketing and comms lead. She will also talk a bit today about projects um and she is responsible for everything you see and hear around Transkribus in the world um like content, website, webinars, and so on together with the comms team. And yeah, I'm here, Michaela, from the board of directors of Transkribus. My responsible area is the go-to-market area. So, everything that goes into the direction of users and customers and the market itself. Um and together with Zara and Elena, we will talk a bit about datasets and projects today. And also take a look into the platform directly. Yeah, I will start as I said with a little introduction. You I assume you all know Transkribus as a platform a a a i powered platform and we design this platform to also yeah simplify the time-consuming work and work in general with historical documents. You know there are a lot of different functions and features and tasks and you can do and work in Transkribus. So the main part is of course the text recognition and transcription. This is the heart of the platform. Then we also have possibility to recognize layouts different layouts like tables, fields and so on. We also have opportunities and possibilities to tag and extract different metadata. Then after your transcription is ready, you can of course also search and publish. We also have some features for this in the platform. We have a nice editor where you can manually work in and then we also have of course the possibility to train different AI models. And you can see this is a lot you can do in Transkribus and so now the the question really is when you want or when you want to do and work with Transkribus and you have maybe a project your project you work on do you really have an overview of the different data collections and models that goes into it? Can you find something at once that you maybe want to find be it a document a specific collection or a model you trained for a new project you are working on? And this can be confusing sometimes and maybe this this sounds familiar to you. There is a lot everywhere to find on the platform different collections. You train different AI models. You have a lot of ground truths you created in different projects and they are across the account. And you maybe really don't have the full picture here. You maybe also work together with a large groups, transcribers, citizen scientists. You have reviewers or editors working maybe alone or together in a team. And yeah, you need to share information via email or other communication channels to to really organize your work together on the platform. And maybe you also run into the challenge that at some point that you train a model or different models and you don't really know what kind of yeah, which pages went into the small model training or which version of of these training data. And maybe you lose track at some point. So this is of course the the picture that I drew here for to explain a bit what our two features are about or what the solution the solution they might offer. And what we are why we built these two features is especially for certain also user types or use cases when you are running a multi-person project of course. You need to have an overview of what's going on in this specific project. You maybe have a lot of people work with more multiple collections and sooner or later you are yeah, someone asked for certain results and you have to show them or reproduce them. This is where our new features are coming into play. Then maybe you want to build on the results you received with Transkribus, so you need some kind of structure, versioning data, and also some kind of audit you can trust. No black box you work in, but really something that is explainable. And maybe you also um run text recognition, but any recognition uh in some kind of way across different teams or team where the team members work on the material different kind of in different roles, and you really need to track who did what, when when, and what the status of these um work looks like. And um this is yeah, a common or this is common practice, and now imagine you can or you have every collection, model, or data set you worked or you need for a project in one place, or you have every train training data that you ever created really in this way that it can be tracked back to the day it was used for a certain training for a model. And this is or these are the reasons or some of the reasons we built these two features. On the one hand, projects to really keep the work organized, so really have everything in one on one in one place, one dashboard, and data sets on the other hand to also keep AI training transparent, to know when and which data uh went into a certain model training. And we will start with looking into data sets now. So I will hand over to Sarah, who will walk us through the data set feature. >> Thank you very much, Michaela. And I'm going to share my screen. Okay. So, we are starting with data sets. As if you have ever trained a model, you know that it's difficult to remember what goes into that model, on which pages you have trained the model, and if you are uh training multiple versions of of your model, it's it becomes even more complicated. And at the same time, it's also difficult to reproduce the models and trace all the versions. We tried to fix these problems with data sets. And now you can create a clean and traceable data sets in Transkribus. So, we are also switching the way in which we think training models. The idea is that every model sits on a data set. So, you're not going to to start a training directly from the collection, even if this is still possible, but we would like to encourage users to first create a data set and then train a model starting from the data set to keep your data cleaner and also to know better know what goes inside the model you are training. Data sets in Transkribus are created sets of pages used to train, validate, and test custom AI models. In the collection, you are doing the work on the on the documents you are creating your uh your ground truth, editing the transcription, correcting them, uh the problem happens when you train a model uh on those pages and then you edit again those pages. So, at that point you don't know anymore on which version of the page I have trained my model or if I want to retrain the model, what was included or what not. Uh but we know I know that we all have some uh workarounds to that. You can write that in description of the model. You can keep a spreadsheet that's set up Transkribus to keep track of the pages included. Uh but it's this is not ideal. So, the missing step between collections and models and training a model is data sets. And the data set should ease this process and also allow you to split more easily your ground truth into the training, validation, and test set. And at the same time keep track all of all the versions and uh add notes, create these data sets. And also another big improvement is that with data sets you can take uh ground truth from different collections. So, also here uh yeah, we all have used different workarounds to have all the ground truth in one collection because at the moment, as you know, it's all before data set it was all only possible to start a training from one collection. So, some users copied the data from one collection to another, others used the shortcuts. Uh but this, as we say, is not ideal. Now, with data sets you can create a data set taking the granted pages from various collections. They are saved in the data set in the version that you selected. So the latest save set version. If you edit that the page later, the data set still keeps the version of when you created the data set. So no changes are applied if you work on the document afterwards. And then on the base of that data set, you can train your AI model. It can be a text recognition model or a layout model. And in the moment in which you train the model, the data set version is frozen, which means that is set in time and this ensure reproducibility. And if you want, you can still continue to work on the data set, add new pages, and create a new version of the data set and then eventually train a new model on the a later version. But this way you keep track of all your versions and all the models you have trained on each version. Why use data set instead of collections? We have already said that a bit because this way you can gather ground truth from various collections. You can uh take a sort of picture of your ground truth at a certain time and then you can work continue working on the documents and then go back to the data set later to train the model. And it also makes easier to compare versions and uh check the improvements of your models. And at the same time, you can also create just one data set and train multiple version different type of types of models on that data set. Yeah, as as you know, now you need to redo the process every time and select the ground truth if you want to select all the granted pages if you want to train a text recognition model. Redo it again if you want to train a baseline model with data set the data set is just one. You know what is inside and you can train on that as many models you want want to with the same type of model with using different settings or multiple type of models if you have created a complete a complete ground truth also with baselines of fields and transcription. This means that we are also switching a bit the way you are used to train models. So the first part is always the same. You in the case of a text recognition model you run a public model or on your documents. You correct the text. You save it as ground truth. But then instead of directly start the training from the collection you first create a a data set and then you start the training directly from the data set. We know that it's we are not used to do that but in the end this switch will come it's easy to manage your uh uh data set models and ground truth that way. So we are sure that you will get used to it sooner and it will ease also our your work. So and here you create the data set which And now let's look inside Transkribus. So I'm going to show you how to create a data set. So we are here. This is ways to create your data set is going here under data set and here you have a list of your data sets, the public ones, which are data sets created by other users and uh that are now public. Uh you can if the users if other users make the the data set public, you can look at the description at the content of the model. In the future, we will also give users the possibility to copy uh a public data set if the owner of the data set set a license that allows that. At the moment, this isn't possible, but we are working on that because we need to work on the licensing to make sure that everyone is aware that their data sets aren't just visible but also reusable by others. So, we will give the choice to decide if they want just to make a the model public and visible or public and visible and reusable. But, by default, the models the data sets are private. You can choose if you want to make it publish and we'll see how to do that later. We have our favorites and then and then the models data sets. This comes from public models whose data sets are set as public. We are here and we are going to create our first data set. We enter the name. So, we can say 19th century and writing and then a short description. Uh we can add the grant type that we have. Text recognition is the It includes grant truth for text recognition and layout analysis, for example, and we can add here tags. These tags are uh customer You can customize your those tags and you can in this case, for example, we can enter the names of the author of the writers on which our ground truth has been created. Uh And we can change hosting, big cancer, and so on. So, it's up to you which tags you want to enter here. Then the language uh You can also add more than one language here and the century. The visibility, the first choice is private, so you can only see your data set or you can switch to public if you're okay with making your data set visible to everyone, which means making the the description, but also the images and the transcriptions visible. And then here below there are the advanced settings. You can also choose the license that you want to to use for your data set and add a your the add a DOI. We created our first data sets and now we are prompted to add a ground truth to our data set. I click add from uh my collection. I select the collection with uh my ground truth. I selected the documents, the pages. I go there and I click add to data sets. Here the user interface gives me two options. One is to create a a new data set, but I already have created data, so I go to select existing data sets and I select the data set where I want to add my ground truth. I can also decide to which set I want to add my data set my pages to. There are three possibilities, train set, validation set and test set. The training set, as you know, consists of the pages on which the model is going to be trained. So, the model will learn on those pages. The validation set assess the accuracy of the model, so it's set aside during the training, is used just to assess the accuracy of the model and prevent and to prevent over fitting. And then we have a test set, which is new in Transkribus. This is an independent set of pages used after the training to provide an unbiased evaluation of your model. So, those are pages not seen at all during the the training. And that you can use to evaluate the model on pages on completely new pages. They are still part of your ground truth, but they're not used to to during the training. Our recommendation is to to assign the pages to the train set and then move them inside the data set. So, I will also show you how to split the pages between the training, validation and test set afterwards. So, let's add our pages to all the our pages to our data set and the here we have our data set and this is the initial version. So, the first version. I can also add a note. for example, version created during the webinar. And here you have the options to edit the metadata, add new material from collection and freeze the freeze the version. So, if you want to freeze this version and create a new one. And there is also the auto lock, so so you can see all the history of this data set. What I wanted also to show you, so for example, yeah, we are here and I all the pages all the 100% of my pages are in the training data. I want to split uh uh the pages between the validation and test data. So, we can do it manually like we select uh the pages and we move them. But for a better for better training and better evaluation, we recommend to use the automatic data set splitter. So, you go there and you decide if you want to split your data set between the training and valid- validation data or between the training, validation and test data. And a good ratio would be assigned 10% to your validation data and 5% to your test data. And automatically the splitter is done but randomly by Transkribus. So, now we have our training data and our validation data and also here our test data. You can at the moment you can create the test data, but the but we are still working on the evaluation features on the test data. So, you can create it is there, but there is no a to currently to measure the character rate on the test data, but we are working on it. So, it's a feature that we will release soon. Okay, we have created our first version. Let's say that uh I've worked on more ground truth and I want it to add I want to add more ground truth. How can I do that? First, I have to decide if they want to create a new version or not. If I want to create a new version, I go here and I click create new version. And automatically, the first version is still visible here. So, I can look at it, but it's frozen and cannot be modified anymore and it cannot uh be canceled. So, be aware that when you create a new version uh or when you train a model the previous version is frozen and it cannot be modified anymore. So, we are here in version two and I now want to add more more data. So, I can select a different collection. I go there and I click add to data set, select this existing data set and add it here. And also here, I can decide to which set I want to add it. And here, I have now my second version and I can add in note additional pages or whatever I want. What I also wanted to show you is that if I try to add pages that are already present uh in my data set uh so Yeah, this page is blank, but just for a test. So, I go to here, add to data set uh and I should receive a message that this item already exists in my data set. So, Transkribus skipped it because this page with that transcription is already present and Transkribus isn't going to duplicate it in my data set. So, there is also this check in the background. Now that we have our data set data set, we are going to train a a model. Training a model is is as before. Just you don't start it from the collection, but you start it from here. So, you click here, train a model. You decide which type of model you want to use. And you start the training. The only difference is that now you are asked to enter a target collection because yeah, every model should have a collection to which they're linked. And here you enter the name, the description, and you go on with the model setup as as usual. And I can also show you that writing uh this model is trained on English documents. Uh And what I wanted to show you is that the the vision between the train the validation and training set is kept. So, I don't need to select it at this stage, but the split between validation and training training and validation set is taken from my data set. And I can start uh the training and wait until it is is finished. [clears throat] If I go back to my data set, we'll now see that version two is now frozen because I used that uh for my training. So, automatically when you use a version to train a model, this is frozen. This is a choice that we made because in only in this way we can ensure the reproducibility of the model. So, you really know on which version you have trained the model and you cannot modify it anymore. So, you can retrain you or someone else can retrain the model the same exact model on the same exact ground truth pages. Yeah, I think I explained the main features. You can cite the data sets the data set if you want and uh if you want to make it public not from the the beginning but at a later point, you can always go there under edit metadata and switch it the visibility from private to public and also update the license and then update the data set and it will then show up under the public data sets. Uh that's all and now Eleni will show will talk about project. I don't know if we have questions or if you want to keep them for the end. >> There are no questions for now. We'll just move on with projects. Thank you, Sarah, for explaining the data sets. This is also something that will come in very handy to already know what data sets are when we are talking about projects. So, we already had the introduction from Michaela about the projects, what they're here for. Let's just repeat that uh the introduction, maybe go a little bit more into depth about projects. So, if you work with Transkribus and you've been using Transkribus for a while, you already know that if you work work starts to scale up, things can get a bit maybe not chaotic, but it's a bit difficult to keep an overview. So, if you are tracking different collections, different documents, trying to remember what model was trained on what collection, it can become a bit tricky to keep the overview. You have unclear permissions, data silos or what we call or it's a bit difficult to keep track of the workflows. And that is why we have the projects. So, you can think of the projects as a higher organizational layer in Transkribus. Um so, it's basically a centralized space where you can manage different assets. So, it's really designed to let you group and structure everything you're working on in one single place. So, what do we mean when we talk about the assets? So, within a single project, you can link a few different thing things, which is of course your collections. So, where the material is that you're working on, the documents that you're editing, that you're transcribing. And also, of course, your models. So, you can also link your models public or privately trained custom trained ones in your projects well. You can also link Transkribus Sites. Um so, Transkribus Sites, just to maybe someone hasn't tried out publishing sites yet, it's a publishing platform where you can easily share your transcribed material online so that everyone can access and search your documents. Uh you can also link those in projects and of course data sets, which Sarah just explained what the data sets are. So, they're curated sets of ground truth data. So, to put it in a different way, um projects are a collaborative workspace that group your collections, models, data sets, and sites together. They all live in this place, basically. And this is to give you a better work structure, but also to make collaboration easier, to make it easier to work with other people on different collections, on all of your work, and give you better visibility and overview of everything that you're actually working on in one collective space. So, just to drive that point home, it is the new organizational layer that can hold all of the assets that you can see here on the side. Now, we already said it can hold all your assets, all your resources in one place, which is the collections, models, data sets, and sites. And you have an overview of the progress across all the collections visible in one single progress bar. You can also add metadata, so tags, for example, what document type you're working on, the time period, also country, for example, the project is happening, also language and other information there as well. You can also control the visibility, so you can keep the projects private only to yourself, but you can also publish them organization-wide, or you can even make them public, for example, for crowdsourcing. And you can also add members to your projects. Again, collaboration is a big part of using projects as well. And you can even give them different roles within the projects. So, let's create the project in Transkribus. Let's see what that looks like. So, when you log in to Transkribus in your in the projects workspace, you can simply start working on a new project by clicking on new project here. Now, we'll enter a title. Let's call this one English say diaries 18th century and then a description. Just copy paste that one in here. So this basically just explains what project this is, what kind of document you have in there, maybe also some more information about the models, etc. You can either choose one of the default cover images. Let's just pick this one or of course you can choose an image of your that you yourself have to either give indication of the material that you're working on. Maybe you have a logo of a project that you're working with. So you can really choose what image you want to upload there as well. And here you can add additional metadata and information about the project where we already saw a preview before. So a country for example, since we or I am sitting in Austria currently, I'll add Austria as a country. The city is Innsbruck where we have our office. So let's add that as well. We can also add an institution if you're working with a specific institution or archive, you can add that here as well. Of course the document types for example, that can then be diary or maybe letters. We have this information here as well. And then let's say from just add a number to 1850 spelled 1850 and then add a language which will in our case be English. Then click on next. And now you can invite team members with either a team plan or an organizational plan. You can search for them with the email address and projects can only be shared again with team members if you are on paid subscription plan, a team plan or an organization plan. Now we will skip this for now. This account is not connected to an organization account because if we would have connected it, it would show all the email addresses, the internal email addresses of our team, and because of data privacy reasons, we don't want that. So, we will skip it for now, but this is where you can add the members to your project. And then you can add the assets and resources to your project. So, let's just start with the collections, and we'll add documents for the beginner webinar, English handwriting. Let's just pick those two for now. We can also add a site or a model. Let's just start off with collections for now, and then we can already create a project. Now, it can happen that not everything is shown right away, but that might just take a bit to load. Let's reload the page, refresh the page, and we should be able to see the assets that we linked. I think we can already see that we have two collections, but it doesn't show. Let's maybe just go to projects and have a look here. So, we can see that the project has been created. We see the name. We see that we have six different pages across two collections with one team member and even the description that we added. But, let's maybe have a look at this example project that we prepared ahead of time. So, we can see here the collections that we linked to the project. One quick thing to just reassure you, adding a collection to the project will not remove or delete the collection from your main collections overview. So, it won't impact any of the material that you edited. It will just link it and group it in the project. There is one thing to keep in mind. Currently, you You only assign one collection to one project at a time. You can at any time add more resources. If you click here on add resource click on collection. What you can see, you cannot add already linked documents to another one. So if it's already in project A, you won't be able to add it to project B additionally or simultaneously. So what we have added now is collections, but you can also add models. So with models, you can add public or private models here, which is also pretty great because I mean it means that everyone collaborating in the project can easily access them and use them then for a text recognition as well. And you can add them again just by clicking here on add resource as well. You can add more models here. Then we have data sets. And here you can add these specific sets of ground truth to your project workspace. This helps you coordinate also model training within a project. And Sarah has already explained how that works, but you can just click here on add data set if there is not one already added. And we'll just add the one that we have just created here in this webinar. Click on add item. And should be able one data set added successfully. Take a bit of time to actually see it, but we are already see it here that it has been added. And the same thing with Transkribus sites. So you can connect your Transkribus sites to your project. This is perfect if you're preparing for the work to be published, if you want to easily access data that has been published from your project workspace. So really everything that you add here is shows up in in the project overview. So you can really keep track of everything that you're working on with this specific project. Multiple collections, different models, data sets all from one central place. Now, we do have some little time. Let's talk about another topic, which is a publication. We can We have already seen a short preview at the overview page in the site. You can decide to keep your project private, which we can see here as well. So, this is a private project, meaning only I have access to it, only I can see it, but it is still a central place where I can have an overview of all of my assets, of my collections, and models, data sets for one specific purpose, for one specific project. But, you can also choose to make this public or to publish it. And there are two different options of how you can do that. One is So, let's click on options and then publish project. To make it visible only to members of your organization. So, this means that everyone from the organization can see the project that you created, or you can make it public and visible to all Transkribus users. If you make it public, that still means that a Transkribus admin will have a look and review the project before it will become visible to all the users. And so, there is a review process involved when publishing it to all Transkribus users. And then you can also choose how users can join your project. So, either they can auto join without your approval, or you will receive a join request, meaning that you can check who is able to who wants to join your project and who is able to, and you can either accept or deny that request as well. So, this These are the options that you have when it comes to publishing the project. And what you can also do is what we also mentioned briefly before is add other users or members to your project that have editing access. And you can do that by clicking on project settings. And then here you have first an overview of the general information, project settings, then again the project information, which you can also change again. And here we have the option to add members to your project. Again, this is not possible right now because this account is not connected to an organizational account. But now you know where you can add members when you're working with a project and you do have an organizational account. What you can also do is give the different members roles. So you can see here I have the role owner. Per project, there can only be one one owner, but you can set the other roles as editor and transcriber, for example. And if these members are already from your organization, they will also inherit access to linked collection in your projects, which is making it even more easier and smoother to to work together, to have a smoother workflow with your colleagues when it comes to working in project in projects. And I think these are the main pieces of information that you need when it comes to starting and working with a project. You can already see here this is actually maybe also nice thing to show. We tested publishing a project earlier today, and you can see here when it comes to publishing a project, you also see the status of pending admin review in your project overview. So that's also maybe a nice piece of information right there. And I think I can hand it over to Michaela now. Before we come to the Q&A. >> I will share a brief outlook with you. We've heard already a bit, but let me just sum it up a bit. So, a little outlook on what we are working on at the moment when it comes to those two features. For the projects, we are working on implementing a bigger task management or task assignment opportunity option here, so that you can together with the team you're working with really hand out page ranges they should work on or specific collections they should work on. You can track progress for different tasks. So, for example, someone only correcting your recognized text. You can set tasks here. You can set the different page ranges they should work on, and you can also track the progress here. So, you know where they are and yeah, when maybe to expect them to be finished. Um there will be some tasks we just have a little overview of what we think of we want to implement here. So, this is only a mock-up, but we're working on this at the moment. And yeah, as I said, you can then also check the different stages of a task. So, not hopefully not handling all these in a different spreadsheet then. So, you can it's really integrated into Transkribus, into your projects with an overview a dashboard where you can look at the progress. >> [snorts] >> And yeah, so you can also yeah, have a better overview and really make it visible on on the stages and the stages in this project. And when it comes to data sets, Sarah already mentioned it, we are working on better options for sharing the data sets or publishing the data sets. So, we will introduce specific tiers that you not only are able to decide if you want to have your data set as a private one or if you want to make it public, but we will also introduce the differentiation between making a data set public with only a view option and on the other hand sharing the data set for others to use as well. So, this is going to be introduced really soon. And then on the other hand, we are looking into multiple options for citing your data set. So, this is mainly important for everyone who wants to or who works in research, for example, and really wants to have a site-able research output for the data set they created during their work. So, we will also introduce more options here so that you can, for example, automatically create a DOI or something else. So, this is what we're looking into the different options, and we will also implement this soon. And then, as always, to to sum this session up with all these improvements, all these new features we are building, we are always on our mission to provide the best tools, the best platform for making history accessible, for working with historical documents. Um you probably already heard in in one of our other webinars about Read Coop, the cooperative behind Transkribus. So, we are more than 250 um co-own or we have more than 250 co-owners um in this cooperative model from public institution to private institutions but also private individuals who co-own our um cooperative and Helene already mentioned we are based in Innsbruck in Austria where also our data center is and our idea or our um mission is also to reinvest everything we earn with um Transkribus to be able to um do good development work to develop the platform further to also integrate new ideas and features that come from the community and the team behind this is um currently around 25 um people strong so a small team but a really dedicated team working on the platform in different roles as well. Here are some of our um members of the co-op and you've heard a lot about datasets and projects today. Please try it out, give us feedback. It's still under development so there's more to come to these features and we are always happy to hear your feedback and all your um also your ideas when it comes to development. Um you can also take a look at our help center. We will publish also um pages on datasets and projects soon where you can read everything um that will hopefully help you work with these two features. And as always if you want to take a look at our webinars you can rewatch them on YouTube. Um you can also take take a look at our website if you want to keep an eye open on upcoming webinars. We will probably um yeah have more webinars after the summer break starting in autumn. And um um yeah, then if you haven't heard it yet, we also have a nice conference in September in at the University of Passau, our Arabic translation Crema user conference every 2 years under the topic not all AI is created equal. We have a nice program, lots of interesting talks, workshops, and round tables. You can of course uh yeah, participate on site where we would be really happy to see many of you on site, but you can also if you like join uh digitally. Um we will stream lots of sessions as well. Yeah, you can also join the conversation if you like. Uh keep an eye open on our socials and I think we have some questions before we end this session. Let me just take a look. >> Um one question was um I mean like you said, Michaela, it uh features are still under development, so we're happy to uh get feedback and hear um what the users would like to do and how they would like to work with the features. Um one question was uh if there is a reason that it's not possible currently to add people or accounts from other organizations to your private projects. >> There is the only reason at the moment that it's technically not possible, but we are working on this already. So, it will be ready soon to also add people from outside the organization. >> [gasps] >> So, we we simply have to implement it from the technical side. And that's not that easy to do, but uh we will introduce this. >> Another question was if it is possible if you are on a team plan which has up to five user seats, If it is possible to add more than five people if they would only do transcription work, so if their role would be transcriber. I guess is the question. I think this was the question of two people in the chat. >> The thing that goes in the same direction that we will open the projects up to also people outside the actual organization or then in this case the team plan. So, this is something we have to implement um we are working on this at the moment. And then it will be possible. >> I don't know if there is anything else. >> I think so far we have answered the question in the chat. Sarah has answered some directly. >> Yeah, if you have any other questions also later, you can always write to us in the help desk to help@transcribus.org where our team is happy to to support you, to answer the questions you have also when it also to yeah, if you have ideas also to collect these ideas, so please write to help@transcribus.org and we are happy to work together with you on questions and ideas, everything you have. >> It seems we have missed a question Um >> Yeah, can >> Ah, yes. >> Question is, can you share credits within a project or does every member need to have a user seat? >> Yeah, at the moment it is impossible to share credits within a a project. So, credits can be shared in teams and or organizational plans, but not within a project. >> Yeah, so at the moment it's only possible Yeah, within the augur plan or team plan you're in. >> [snorts] >> Not in the project. Then we are right in time. Thank you for joining us today for this webinar on projects and datasets. We're wishing you a great Wednesday evening. Have a nice week. And hope to see you soon or also hear from you through our channels if you like.