Hello everyone! We've been slow on news posting in these recent.... weeks at this site, I am aware of it. We've been relying on your news posts and announcements to keep this place alive. So, if you have something interesting to announce and spread the word, please do it by posting at our Community News Forum. But, this doesn't mean the death of this place. I am very passionate about this place and I wouldn't be crazy to just throw it away out of no where.The current situation happens because of a bet for a better future. You know? Sometimes big steps ahead demands big sacrifices.
The current sacrifice is directly related to my doctorate, which I'm having a hard time to finish it. While it is obviously more important for my real life, it will certainly have a great influence at this site as well, mostly at our future modeling tools. Doctorate candidates are obligated to research things that were not covered before, so if you have a plan to make one doctorate, PhD, etc, you should be very careful at revealing your research plans around. It's possible that someone at the other side of the world publish the same thing that you are researching, which may possibly ruin your situation. So, I've been quiet about my research for all these years, at least in the internet. But right now, I'm forced to finish it in a couple of months and, to be very honest, I don't think anyone will be able to finish it in such time, so I can talk a little about it.
As a visual computing scientist, I think it's amusing how most techniques (if not all) applied in my area to handle visual objects or solve problems with it doesn't really know what kind of objects they are dealing with. It's actually kinda ironic, in a certain way. Many of these techniques are based on local approaches. Imagine someone who is blind and deaf at the same time trying to know a place or trying to figure out how to go somewhere. While this person could use their hands and body to know what is around it, he can't know what is going on in the whole scene and use it in his advantage. For you guys at PPM, samples of such approach is the autonormals detection at the Voxel Section Editor III, image filters and even the 3D reconstruction techniques in the 3D Modelizer feature of VXLSE III. Other techniques may subdivide the space into smaller parts and treat them as local neighbors. I.e.: Octrees, Wavelets, skeletons and most if not all known hierarquical structures, etc. It's smarter, but you are still influenced by how your object is sampled in first place, which may generate noise in the solution.
You don't need to be acquainted to computer graphics to know that humans doesn't view objects like that, as we pay much more attention to its contents rather than how it is discretized, since there are far too many points for our eyes, isn't it? Objects are subdivided independently of the space metric, but mostly based on our semantic knowledge of them. And that's how we understand objects, languages and any kind of information in the end of the day. And that's what I'm trying to approach with my doctorate research, a graph language that allows me to understand objects and that is far more insensible to noises (bad sampling) than any existing graphic data structure out there.
By understanding objects, we could do a lot of things with them:
- Reconstruct them (zoom in, increase sample resolution, polygonize graphics, etc)
- Parametrize them (so, you can lately add some textures, or you could also use it to compare different objects and move information from one to another)
- Recognize them (and catalog them without using existing heavy brute force methods, and then using existing objects to add details to new objects of the kind... or even finding out how to automatically edit certain objects to make them fit into new contexts).
- Distinguish objects and different contexts where they belong, as well as allowing some of them to be auto-completed.
- Pathfinding (it could even be used in RTS games to fix some of the existing pathfinding problems, allow AI to know the best places where to set its defensive buildings or to find the weakest spots of the enemy and send its army in a path where it could exploit it).
It's worth mentioning that I don't have any of these things working at the moment and, in the best of the situations, I may have reconstruction working by the end of my thesis restricted to 2D objects. Coding my proposal is complicated because I have to deal with a bunch of complicated computer graphics and compiler problems at once. And that's what is cracking me so much. But still, this is the first step to Skynet a better understanding of visual objects as we know and a huge set of modeling tools that could make use of it.
The current sacrifice is directly related to my doctorate, which I'm having a hard time to finish it. While it is obviously more important for my real life, it will certainly have a great influence at this site as well, mostly at our future modeling tools. Doctorate candidates are obligated to research things that were not covered before, so if you have a plan to make one doctorate, PhD, etc, you should be very careful at revealing your research plans around. It's possible that someone at the other side of the world publish the same thing that you are researching, which may possibly ruin your situation. So, I've been quiet about my research for all these years, at least in the internet. But right now, I'm forced to finish it in a couple of months and, to be very honest, I don't think anyone will be able to finish it in such time, so I can talk a little about it.
As a visual computing scientist, I think it's amusing how most techniques (if not all) applied in my area to handle visual objects or solve problems with it doesn't really know what kind of objects they are dealing with. It's actually kinda ironic, in a certain way. Many of these techniques are based on local approaches. Imagine someone who is blind and deaf at the same time trying to know a place or trying to figure out how to go somewhere. While this person could use their hands and body to know what is around it, he can't know what is going on in the whole scene and use it in his advantage. For you guys at PPM, samples of such approach is the autonormals detection at the Voxel Section Editor III, image filters and even the 3D reconstruction techniques in the 3D Modelizer feature of VXLSE III. Other techniques may subdivide the space into smaller parts and treat them as local neighbors. I.e.: Octrees, Wavelets, skeletons and most if not all known hierarquical structures, etc. It's smarter, but you are still influenced by how your object is sampled in first place, which may generate noise in the solution.
You don't need to be acquainted to computer graphics to know that humans doesn't view objects like that, as we pay much more attention to its contents rather than how it is discretized, since there are far too many points for our eyes, isn't it? Objects are subdivided independently of the space metric, but mostly based on our semantic knowledge of them. And that's how we understand objects, languages and any kind of information in the end of the day. And that's what I'm trying to approach with my doctorate research, a graph language that allows me to understand objects and that is far more insensible to noises (bad sampling) than any existing graphic data structure out there.
By understanding objects, we could do a lot of things with them:
- Reconstruct them (zoom in, increase sample resolution, polygonize graphics, etc)
- Parametrize them (so, you can lately add some textures, or you could also use it to compare different objects and move information from one to another)
- Recognize them (and catalog them without using existing heavy brute force methods, and then using existing objects to add details to new objects of the kind... or even finding out how to automatically edit certain objects to make them fit into new contexts).
- Distinguish objects and different contexts where they belong, as well as allowing some of them to be auto-completed.
- Pathfinding (it could even be used in RTS games to fix some of the existing pathfinding problems, allow AI to know the best places where to set its defensive buildings or to find the weakest spots of the enemy and send its army in a path where it could exploit it).
It's worth mentioning that I don't have any of these things working at the moment and, in the best of the situations, I may have reconstruction working by the end of my thesis restricted to 2D objects. Coding my proposal is complicated because I have to deal with a bunch of complicated computer graphics and compiler problems at once. And that's what is cracking me so much. But still, this is the first step to Skynet a better understanding of visual objects as we know and a huge set of modeling tools that could make use of it.