Jump to content

  • Log In with Google      Sign In   
  • Create Account


jefferytitan

Member Since 03 Nov 2009
Offline Last Active Sep 02 2014 05:17 AM

Topics I've Started

Any interest in reverse engineering .unitypackage files?

31 July 2014 - 04:00 PM

I'm currently doing a little home project that involves reverse engineering the .unitypackage file format. I have a lot of free Asset Store assets downloaded, and find the built-in categorisation system useless for finding what I need. Therefore I'm making a .NET tool that runs through my asset store folder and stores package names, paths, filenames and preview images in a SQL database so I can do full-strength searching. This would allow me to (for example) find all *.anim files, or all model formats that contain "car" in the path.

 

It likely wouldn't cover all facets of the file format, just the ones I need for my project. Anybody interested in an article when I've finished my little project?

 

JT

 


Neural networks fundamentally flawed?

27 May 2014 - 08:57 PM

Hi all,

 

To prefix this discussion, I know NNs aren't used much in games, presumably because they can have unpredictable results and be hard to tune. The below is an article too juicy for me not to want to discuss it somewhere, and here seemed an okay place to bring it up.

 

Researchers found some interesting things about the stability and continuity of the mapping between inputs and output with NNs which (for me) cast some pretty big doubts on their overall usefulness for most purposes.

http://www.i-programmer.info/news/105-artificial-intelligence/7352-the-flaw-lurking-in-every-deep-neural-net.html

 

As far as we can tell, these are issues that don't occur (or occur much less frequently) in organic brains. A few theories on my part about this, I'd be interested to hear other perspectives:

 

  • Our neural net training algorithms are faulty, e.g. very different to in nature.
  • The simple layered approach is faulty, e.g. in a real brain signals can bounce around many times before producing an output, rather than hit each neuron exactly once.
  • Models neglect the time factor, e.g. we have a continuous stream of input rather than a single input, we may take time to make a judgement.
  • Our ability to act is a crucial factor for learning, e.g. we can interact with reality to confirm and clarify theories.

I welcome your thoughts.

 

JT

 

Edit: The tone of the article may be causing confusion, so I found a link to the actual paper

http://cs.nyu.edu/~zaremba/docs/understanding.pdf


Graphics baseline for a good-looking PC game?

11 May 2014 - 05:14 PM

I was thinking about a variety of games that I've played. It can be hard (for me) to tell which graphical techniques they use. I would never aim to write a AAA killer (due to the effort/reward ratio), but what sort of techniques are expected for a game not to look dated? A few examples:

  • Textures - simple, with normals, relief-mapped, procedurally generated, sub-surface scattering?
  • Geometry - number of polys, tessellation?
  • Lighting/shadows - forward, deferred, soft shadows, SSAO, global illumination?
  • Effects - HDR, bloom, FSAA, volumetric fog, god rays?
  • Sky - skybox, animated, day/night, weather, seasons?

What would be your baseline must-have for a FPS that won't turn off paying consumers?


Function with specific characteristics

11 March 2014 - 04:15 PM

Hi all,

 

I'm working on something horror-themed, possibly for a portfolio. I have a visual effect which can be applied a certain amount, let's say a floating point number from 0 to 100 for the sake of argument. In keeping with the horror theme I want the effect to progress in a somewhat jerky unpredictable manner with pauses. Any assistance with choosing an appropriate function would be appreciated. Some desired characteristics are below:

 

  • Monotonic
  • Some pseudo-random component
  • Has pauses (plateaux)
  • Has some sense of momentum (e.g. derivative has some pattern, not just spikes)

Some approaches that I considered which don't quite make the grade:

  • A random walk with another function as the envelope (no pattern to derivative, timing too predictable)
  • Some monotonic function plus noise (timing too predictable, may not be monotonic)

Maybe I'll have to go piecewise, but I hope not.

 

Any thoughts would be appreciated!

 

JT


Making smart-seeming NPCs

16 September 2012 - 12:57 AM

Hi all,

I was just pondering on what aspects of a humanoid NPC make them seem smart/dumb. Personally I find that the moment an NPC loses believeability I stop treating them like a character and start treating them like an obstacle, a puzzle or a resource to be mined. It also becomes frustrating because I tend to lose my suspension of disbelief. What aspects do you think make the most difference? I have a few examples below:
  • Bad pathfinding
  • Lack obvious abilities, e.g. climbing stairs or fighting
  • "Forget" that they know you
  • Completely predictable
  • Repeatedly fall for the same moves
  • Bad/no gaze tracking, including talking to where you were when you walk away
  • Lack of context, e.g. will talk to you about the weather when in danger
  • Keep to script even when it makes no sense, e.g. talking to a dead comrade
Look forward to hearing your responses.

JT

PARTNERS