/sci/ - Science & Math » Thread #12601508

258KiB, 800x1200, aliencard.png

View Same Google iqdb SauceNAO

Research Topics in Computer Animation

Anonymous Tue 19 Jan 19:04:29 2021 No.12601508 View Reply Original Report

Quoted By: >>12601635 >>12607931

Undergrad cs major, one of my classes is going to be regarding writing a paper relating to scientific principles underlying humanoid character modeling and animation. I seriously have no clue on what to do or how to research what to do for this topic. Anyone know any ideas that seem interesting, i'd like them to relate to artificial intelligence if possible

Anonymous

Anonymous Tue 19 Jan 2021 19:26:45 No.12601635 Report

Quoted By:

>>12601508
Yeah, OP. The most important scientific principle is the positivist inquiry. From an observation of the world, you make a hypothesis, you test it and write down the conclusions which should be revised and replicated by other scientists.

Anonymous

View Same Google iqdb SauceNAO 1601881346966.jpg, 267KiB, 842x1094

Anonymous Tue 19 Jan 2021 19:35:34 No.12601674 Report

Quoted By: >>12601702

the ideas underpinning modern 3d animation are straight-forward, but there is a lot of effort that goes into generating the data that gets processed. fortunately there are good tools (e.g. blender) for generating that data.

3d animation requires a solid grasp of linear algebra is a must. if you aren't good at it, this is also a good time to pick some of it up.

i'll give you a high level overview:
computer graphics works with 4d homogeneous coordinates
geometry is represented by mesh data structures (verts, edges, faces)
geometry is defined w.r.t a local coordinate system basis
the basis (and the geometry/verts) is posed with a 4d matrix

deformable models require a tiny bit more sophistication:
bones can be defined as coordinate system bases w.r.t the geometry local basis
each bone (and the geometry/verts) is independently transformed by its own 4d matrix
the weighted average of all the points is taken to compute the final pose of the verts
the weights are defined when skinning the model
this is called Skeletal Subspace Deformation

to fix problems with collapse and pinching with SSD, modify the underlying geometry before deformation:
for each mesh, define multiple "shape keys" (versions of verts, keep the same edge and face topology), such that different versions transform better under SSD
take a weighted average to combine the shape keys
then apply SSD
this is called Shape Key Deformation

shape key weights can be defined using "drivers" that are functions that relate pose parameters to shape key weight. In this way, an elbow can change shape keys as it bends so it looks good at any pose.

animation is really a matter of defining curves as a function of time, evaluating the curves at different moments in time, and then applying the result to animation parameters, most notably pose.

i've written a library that extracts geometry, bones, weights, and animation data from blender and imports them into matlab (ugh).

Anonymous

Anonymous Tue 19 Jan 2021 19:40:28 No.12601702 Report

Quoted By: >>12601811

>>12601674
i should add all of this doesn't render you anything, it just gets you the geometry needed to throw at a rendered. rendering, be it raster graphics of raytracing, are other (longer) stories

Anonymous

Anonymous Tue 19 Jan 2021 19:49:58 No.12601744 Report

Quoted By:

also, sites like heroforge are using interesting technology that is.

if you want an AI bent to it so you can, say, create novel people models, you can form a linear map between geometry/verts with a standardized topology, and biometric parameters etc using Singular Value Decomposition. Invert the map you get, and you can dial in new biometric information (or randomize it) apply the inverse map, and now you have a new person.

standardized topologies are fun because they allow you to math on people's bodies. you can also transfer bones, weights, and texture coordinates this way between people.

none of this is new, but it's fun to mess with. see heroforge.com for a good practical example

Anonymous

View Same Google iqdb SauceNAO super-mario-bros.jpg, 431KiB, 630x630

Anonymous Tue 19 Jan 2021 20:02:32 No.12601811 Report

Quoted By:

>>12601702
This is really interesting, looks like I need to review some stuff. Do you know of any papers I could read or any general ideas that would make for a good research paper? My entire course is going to be dependent on it so I want to come up with an interesting one.

Anonymous

View Same Google iqdb SauceNAO 6791586.png, 162KiB, 640x352

Anonymous Tue 19 Jan 2021 22:10:17 No.12602272 Report

Quoted By: >>12602453

pick up Shirley. it's a standard graphics text for basic stuff, but be warned that early editions have serious errata.

i love graphics people because they make everything open access. this paper is where i got a lot of the ideas i shared with you from:
https://grail.cs.washington.edu/projects/digital-human/pub/allen03space.html

there's a lot of practical and creative applications with this. give me some idea of your background.

personally i'm doing research where this is key to doing post processing on images of people in unconstrained imaging scenarios.

Anonymous

Anonymous Tue 19 Jan 2021 22:33:42 No.12602340 Report

Quoted By: >>12602453

Two minute papers with Karoly Zsolnai Feher has a lot of interesting stuff on youtube specifically about ML applied to computer graphics research.

Anonymous

Anonymous Tue 19 Jan 2021 22:43:10 No.12602372 Report

Quoted By: >>12602405 >>12602453

Yes 3d mesh reconstruction from single shot rgb(d?) image. this kind of thing https://arxiv.org/abs/2101.05356. preferably using gan *and* neural radiance feilds (nerf). nerf is pretty new, and seems promising. What do you think?

Anonymous

Anonymous Tue 19 Jan 2021 22:54:53 No.12602405 Report

Quoted By:

>>12602372
Here's another example https://arxiv.org/abs/2101.02000. face reconstruction is very popular and useful. these two papers are from this month

Anonymous

View Same Google iqdb SauceNAO image_2021-01-19_150835.png, 211KiB, 941x380

Anonymous Tue 19 Jan 2021 23:09:39 No.12602453 Report

Quoted By: >>12602878

>>12602372
>https://arxiv.org/abs/2101.05356

Looks cool, but don't think I can do something like that in 1 semester.

>>12602340
I'll check it out now

>>12602272
I'm doing a math heavy cs degree, I've taken some probability courses, linear algebra, and differential equations, also have some knowledge in a.i.
Pic related, the paper has to be a technical report and a project about one of these topics.

Anonymous

Anonymous Wed 20 Jan 2021 00:55:54 No.12602878 Report

Quoted By: >>12604313

>>12602453
something that might be useful for advanced humanoid animation is knowledge of inverse kinematics and relatedly solving ill-conditioned linear problems by regularized iterative methods

Anonymous

Anonymous Wed 20 Jan 2021 08:45:24 No.12604313 Report

Quoted By: >>12604325

>>12602878
Those videos were really interesting! It really opened my mind to the possibilities I have for this course. I think I want to do something relating to a human ai interacting with the environment, maybe something like two ai's playing a sport against each other.

Anonymous

Anonymous Wed 20 Jan 2021 08:53:58 No.12604325 Report

Quoted By: >>12605379

>>12604313
I think I'll try to model the player batting in a game called cricket, I don't think it's been done before so it'll be interesting.

Anonymous

View Same Google iqdb SauceNAO poop.png, 158KiB, 334x609

Anonymous Wed 20 Jan 2021 15:51:23 No.12605379 Report

Quoted By: >>12605414

>>12604325
i've been wanting to play around with evolving virtual life and its locomotion. this might give you some ideas on ai's application to animation

https://www.youtube.com/watch?v=JBgG_VSP7f8
https://www.youtube.com/watch?v=pgaEE27nsQw
https://www.youtube.com/watch?v=hx_bgoTF7bs

Anonymous

Anonymous Wed 20 Jan 2021 16:02:10 No.12605414 Report

Quoted By: >>12606970

>>12605379
These were interesting applications, do you know of any good resources to learn further about the topic?

Anonymous

Anonymous Wed 20 Jan 2021 23:08:33 No.12606970 Report

Quoted By: >>12607754

>>12605414
just search the authors in the video.

another idea might be to try to classify different human poses. the more vr becomes a thing, ai will need to be able to recognize what people are doing to react appropriately

Anonymous

Anonymous Thu 21 Jan 2021 02:37:05 No.12607754 Report

Quoted By: >>12609733

>>12606970
Thanks anon, if I ever end up making something cool/interesting i'll share to the site.

Anonymous

Anonymous Thu 21 Jan 2021 03:20:50 No.12607931 Report

Quoted By:

>>12601508
Machine learning + animation frame interpolation will lead to general AI

Anonymous

Anonymous Thu 21 Jan 2021 17:16:47 No.12609733 Report

Quoted By:

>>12607754
cool, hope to see something neat in a few months

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches