3 ms·
> From a computer vision perspective, I think most of that is fairly easy No, not at all. The hard part is to do computer vision to understand what you can do
by Jensson 5y ago
> From a computer vision perspective, I think most of that is fairly easy
No, not at all. The hard part is to do computer vision to understand what you can do with objects, not just convert object representation of images into string texts. For example, can I move this object away to get an object under it? Can I place an object on this object without it toppling over? Is this object dirty and moving it will smear things all over? If I topple this object, will it break? How much pressure can I apply to this object without breaking?
Those things are necessary to have an intelligent agent act in a room, every animal can do it, and our software models are nowhere near good enough to solve them.
You need the computer vision program to also have a physics component so it can identify strain, viscosity, stability etc on objects, and not just a string lookup table it needs to understand that naturally like humans as the same object category can have vastly different properties identified by looks.
- fault1 5y agoI agree with all that (though I don't really necessarily associate any of that with computer vision, perhaps it's my physics bias). Having good state estimation, kinematic, and dynamics models of real world objects is something that is very mature in controlled environments, but not very mature in other environments.
- pdimitar 5y agoHm? Do we have robots that can pick eggs without squishing them, for example? I vaguely remember reading something like this years ago and it was impressive.
- xaedes 5y agoYes, "soft robotic grippers" can do it.
- Jensson 5y agoI think you might over-estimate what the current computer vision models are capable of. They are very good at recognizing what class an object belongs to, but they aren't very good at extracting data about the object based on the image. Extracting data of objects from images is image recognition, and humans and most animals relies heavily on vision to get data about objects.
- p1esk 5y agothey aren't very good at extracting data about the object based on the image I’d say they are pretty good: https://towardsdatascience.com/a-guide-to-image-captioning-e9fd5517f350 https://towardsdatascience.com/a-guide-to-image-captioning-e...