 My name is Martina Polik. I'm a PhD student and research assistant at the Cyprus Institute. I will talk about the gravitate project today, which is a project that was funded by the European Commission in 2015. I am it's a multidisciplinary project with several partners across Europe and in Israel. I myself, I'm, as I said, part of the Cyprus Institute. Then we have the University of Amsterdam, IT Innovation, which is in the UK, the British Museum, IMAATI, which is a group of researchers from the National Italian Research Council in Genoa. And then the two partners from Israel, the Ticnion and the University of Haifa. So before I will go into more detail into the talk, I want to give you a brief overview. So beginning, I will talk about the ecological challenges that we are trying to address as a project. And then I will talk a little bit about the research approach. And then finally, I give you a preview of the platform as we are developing. Because the project's goal is to develop a platform for the study of vice-versa collections of cultural objects. So the starting point of our project was the Salamis collection. Until then we have, or since then, we have incorporated more datasets. But the starting point was the Salamis collection, which is a group of fragments, about 250, of votive terracotta statues dating to the 7th and 6th century BC. And they were unearthed in Cyprus and the site of Salamis, so in the south-eastern coast, a British mission in the 19th century. And from the very beginning, there were a few problems in the interpretation. The collection wasn't studied immediately, but it was already from the beginning a bit difficult because it wasn't obvious how many statues there were and how many types. So it was a bit difficult to interpret the context. And this is very common with other ecological collections as well, because we have to deal with broken and incomplete pieces and we have to deal with a large amount of data. So this is our first challenge. And the second one is dispersion, what I mentioned before. Because the collection was distributed over several countries. A big portion is in the UK right now, in the British Museum, and then there's also some in Yosemilian, Fitzwilliam, and then some that are still in Cyprus. And also here we encounter a few challenges, one of which is access and rights. So if you want to study this collection, you have to travel to all these places in person and look at the pieces. And then of course you have to deal with heterogeneous metadata structures and content. As in before you go inside, I'm usually you look at the museum record to get an idea of what is there. But as you know, not all, the museum records are the same. They might differ in their level of detail in the descriptions. They might use different definitions and terminologies and even something quite straightforward as a measurement might not be necessarily. For example, in that case, we don't really know the orientation of this fragment. And depending on how you hold it, you will get a different preserved type. So if you want to get a real idea, you have to really physically go there and touch the object yourself. So how does this relate to our project? We have three key words in the project which is reunification, re-association and re-assembly which are our three aims. So in terms of the dispersed collection, we speak of reunification because within the project, we try to collect all the material and put them together in a virtual repository and bring them all on the same level in terms of the metadata. Then the problem of the large amount of data, we try to develop some tools that allow you to store it quickly and efficiently through all the available data so you're able to make re-associations between the different pieces. For instance, find two pieces that once belong to the same statue or make stylistic connections, things like that. And in the end, the problem with the incomplete artifacts, we try to address that by developing a tool for the semi-automatic re-assembly of pieces always in a digital way, of course. So how did we do that? There was first at the beginning, there was a phase of data collection and preparation. This meant the 3D digitization of every single piece as well as the collection of all the semantic data available from the museum records. And then there was a phase of data enrichment both in a semantic and a geometric way. And then all this information was put together in the virtual repository. So this means containing the 3D models and photographs and all the semantic data structured in the same way in Psyduck CRM. And finally, also the gravitate data meaning the geometric properties that were calculated in the previous phase. And finally, the development of the platform itself, meaning a search engine that allows you to access this virtual repository and that allows you to do semantic and geometric searches. Then 3D visualization and analysis tools, 3D reassembly tools and tools for documentation and metadata enrichment. So for the 3D digitization, we used different kinds of laser scanners and photogrammetry and we were trying to be very careful in the 3D models and produce very high quality models. For example, in terms of the color information, we only used color calibrated pictures. For the data enrichment for the semantic part, that means we used natural language processing and graph matching. So basically these are techniques that allow you to draw more information out of textual descriptions so you can feed that into your search engine and get better results. And for the geometric part, that meant that some properties were calculated, for example, the average thickness or the characterizations of patterns on the fragments, for instance, lines or flower patterns, things like that. So this is now the first picture of the platform itself of the search engine. The search engine is largely based on a project that is being carried out at the British Museum called Research Space, which we have modified a bit to our needs because they work only with semantics, whereas we use both semantics and 3D, so it's a combination. So what you can do there is you can compose queries based on specific elements that those of you who work with SIDOC will recognize. So thing, actor, place, time and properties. So for instance, in the example that you see on the screen, I looked for a thing that is from a specific place, which is Cyprus, and then and a thing that is or has a beard. So basically that means I look for beards from Salamis and I get all the beards from the Salamis collection as an outcome. And then you can further narrow down your results using filters. And of course you can also look at the single pieces in more detail and simply click on it and then you get a separate field that shows you all the semantic information associated with that fragment. And then you can also look at the 3D preview of the fragment, or you can conduct similarity searches both in semantic and geometric. This is how the geometric similarity search would look like. So you have always has a starting point fragment you're interested in. So in that case, the spear, I don't know if you can recognize it. And then you can look for similar fragments. For example, a similar color, similar thickness, similar pattern in order to make three associations. And once you have made a selection of things you're interested in, you can look at the high resolution 3D models of the objects. So here we have just some standard for DV visualization tools. I don't know how familiar you are with them. So that just means that you can zoom in, zoom out, rotate and measure, change the light, all these kinds of things. Pretty standard. The only thing that's quite new or that's innovative is that you always have the connection to the semantic data. So it's not just the geometry, you always have also all the semantic metadata displayed at the same time at the bottom. So here you can drop down and see all the information disconnected to the object. And you can compare objects at the same time as many as you want. And you can also look at some of these geometric properties that were calculated in the previous phase. For example, the mean curvature. This helps you to highlight small variations in the surface. For example, this is useful if you want to study production techniques because it highlights, for example, all the tiny lines that are created when something is made on a wheel, for example. Then there's the possibility to document everything that you do and to enrich the metadata. And there are two different ways to do that. One is annotation, which is essentially 3D annotation. So you select an area on the fragment you're interested in and an example someone has selected, the iris for example, and then you just tag it with a term from a predefined vocabulary that is also within the platform. And then you save it and it's again saved in the back end, which means that it's available in the searches and makes your work more efficient and faster. And the other way to enrich the metadata is through this scheme at the beginning. So through Psyduck, essentially, that you make. So these are observations that aren't directly linked to the geometry, but can be about other things, for example, about the type of the object. Is this a statue? Is this a beer? Is it a face? Or is it not from Salamis? And things like that. And all of this is documented as in you have to state how you make this observation and when, so everything is transparent. And the last bit, the reassembly. So this is an algorithm that works with pairs of fragments that you give it beforehand. So maybe before you have found two pieces that you think they could match, you give it to the algorithm, it runs and it gives you as an output different aligned pairs. And then you can say, okay, I like this alignment or I don't, and if you like it, it will be safe as one entity and again available afterwards. So to sum up, I just wanted to give you or leave you with a few major points. The first one is that within the project, we always try to combine the geometric and the semantic approach, which is usually quite separate. And so in every step, we try to have both things available and to never disconnect them. Then we try to create an environment that allows you to reason and to explore because it allows you to do essentially the entire pipeline that you would do normally if you study an ecological collection. So you have the collection, you look for the pieces that you want to study, it gives you the tools to actually study them and then also to put the new information into the system and make it available for others or for yourself for later research. And this means that this whole system is not static, it's very dynamic and you can continue to use it and put more information in. So if you have more interest in the project, you can always visit our website or also contact us. Thank you very much.