gdalgorithms-list

[Algorithms] vrefresh jumpiness

[Madoc E. <tm...@ti...>]

I've noticed that whenever framerates drop below the monitor refresh rate
renderings look they're jumping back and forth, it basically looks like
there are more offset images overlayed. The problem is most noticeable with
vsync enabled. This problem is visible in all programs and hardware I've
seen and it's _really ugly_. I'm puzzled that I've never seen it mentioned.
I might even have thought that it's some optical illusion that I only am
subject to if it weren't that all my collegues agree that it's very
disturbing (something in the air at the office?).

Anyway, I can't quite understand why it looks as bad as it does. I guess
It's just to do with the way we perceive the monitor's flickering image and
there's just nothing to do about it, I can't seem to improve it. Has anyoene
ever tried to correct it with any success? Am I missing some really obvious
reason why it can't be?

Cheers,
Madoc

Re: [Algorithms] vrefresh jumpiness

[Nils P. <np...@in...>]

Madoc Evans wrote:

>I've noticed that whenever framerates drop below the monitor refresh rate
>renderings look they're jumping back and forth, it basically looks like
>there are more offset images overlayed. The problem is most noticeable with
>vsync enabled. This problem is visible in all programs and hardware I've
>seen and it's _really ugly_. I'm puzzled that I've never seen it mentioned.
>  
>

I think it's slightly offtopic.. I have no algorithm to offer, but an 
explanation:

I absolutely know what you're talking about. I guess most console 
developers also know what you're talking about because they get very 
sensitive to graphics that are not 100% in sync with the vertical 
refresh over the years.

You start to see "ghost images". My argument is, that from watching TV 
the eye is conditioned to see 50/60 frames at the same refresh-rate, 
always 100% in sync. The eye will start to see these kinds of illusions 
when framerate and refresh-rate differs.

If you go into a cinema and watch a movie with 24 or 25 fps you don't 
see the ghost images. On the other hand if you watch a 25fps movie on a 
50hz tv-screen you'll see them. The refresh-rate is much worse in the 
cinema , but your eye/brain is not conditioned for this framerate, thus 
it does not expect the high framerate and can compensate better.

PC players will most probably not notice it as hard as you do, and 
europeans will not notice it because we have a lot of frame-rate 
converted ntsc-movies on the tv (those jerk horrible if you look at them).

  Nils

Re: [Algorithms] vrefresh jumpiness

[Thatcher U. <tu...@tu...>]

On Mar 25, 2004 at 05:07 +0100, Nils Pipenbrinck wrote:
> 
> If you go into a cinema and watch a movie with 24 or 25 fps you don't 
> see the ghost images.

You should, movies are unbelievably bad in this respect.  Pay
attention during any panning shot.  Movies have half-frame-rate-itis
all the time, since they always flash the same frame twice.

The best explanation of the double-image artifact I've heard is that
your eye tries to smoothly track moving objects.  In the case of 30fps
update on a 60Hz screen, with an object moving at a velocity v, your
eye tries to track the object at velocity v.  At 60fps, the screen
would show:

p0 + v*0/60
p0 + v*1/60
p0 + v*2/60
p0 + v*3/60
p0 + v*4/60
p0 + v*5/60
...

Now if your eye is tracking smoothly to compensate for the object's
velocity, the object's image will hit your retina in the same spot
each frame, and create the illusion of a single object moving at
constant speed.  But at 30fps, the screen shows:

p0 + v*0/60
p0 + v*0/60
p0 + v*2/60
p0 + v*2/60
p0 + v*4/60
p0 + v*4/60
...

In this case, the object image will hit your retina in two alternating
spots, and you percieve a double image.

You can run your own experiments to explore this.  A small white
circle moving fairly quickly on a black background works pretty well.
A CRT shows it better than an LCD, but it's still noticable at 30fps
with an LCD.

At 20fps you should be able to see a triple image; at 15fps you may
even see a quadruple image.  It might help to be in a dark room.  Also
take note of your monitor refresh rate; you can still see this effect
at e.g. 72Hz, but you'll need to display 36fps, 24fps, etc.  I.e. this
effect has little or nothing to do with your eye being accustomed to a
particular refresh rate.

-- 
Thatcher Ulrich
http://tulrich.com

Re: [Algorithms] vrefresh jumpiness

[Stephen J B. <sj...@li...>]

Thatcher Ulrich wrote:
> On Mar 25, 2004 at 05:07 +0100, Nils Pipenbrinck wrote:
> 
>>If you go into a cinema and watch a movie with 24 or 25 fps you don't 
>>see the ghost images.
> 
> 
> You should, movies are unbelievably bad in this respect.  Pay
> attention during any panning shot.  Movies have half-frame-rate-itis
> all the time, since they always flash the same frame twice.

You are both right.  Almost certainly, one of you sees one thing, the
other sees something different.

> The best explanation of the double-image artifact I've heard is that
> your eye tries to smoothly track moving objects.  In the case of 30fps
> update on a 60Hz screen, with an object moving at a velocity v, your
> eye tries to track the object at velocity v.

The better way to state this is to imagine the canonical caveman chucking
a rock at a rabbit that he fancies for lunch.  As the rabbit runs behind
a tree, he loses sight of it momentarily - and if his brain didn't deal
with that, he'd be unable to hit the darned thing.  So after enough
generations of hungry cavement, we've evolved the ability to interpolate
the missing data when the object we're tracking vanishes momentarily - and
this is what makes a series of still images in movies and TV appear to be
moving.

I suspect that the lack of interest of some animals in watching TV relates
to their failure to have evolved that mechanism.  For them, TV is just a
slide-show.

OK - so what are the implications of this?

If you plot a graph of position against time of an object being updated
60 times per second on a 60Hz CRT, the points lie on a nice straight,
diagonal line - and our brains have no trouble interpolating the
positions using the million-year-old rabbit-went-behind-a-tree mechanism.

But if you only update the position 30 times a second - but refresh the
screen at 60Hz and plot a time-versus-position graph, it looks like a
staircase - right?

There are two ways to draw a conclusion about the position of the object
between frames from that data.

1) Fit a wavy line or an actual square-edged staircase to those points.

2) Draw two nice straight diagonal lines - one that goes through
    the odd numbered points - and another parallel line that goes
    through the even numbered points.

99% of humans unconsciously do the latter because rabbits can't change
their velocity at 30Hz.  We just havn't evolved to interpolate that
kind of motion data.  So we HAVE to take the two-straight-lines view of
the world and instead of seeing one object moving jerkily - we see TWO
objects moving along perfectly smoothly right next to each other.

Hence the double-imaging.

However as you drop the update rate, our brains 'snap' to seeing
the wavy-line/jerky motion version of events because the motion fits
better to our mental model of a running prey animal.  This seems to happen
somewhere below 30Hz and above 10Hz (for 99% of people - there are always
a few outliers on that bell-curve).

Cinemas are running somewhere in the middle of that range. 24Hz.

That's why *SOME* people see double-imaging in cinemas - and some people
see jerky/blurry motion instead.

On a CRT, if you run the CRT at 60Hz and repaint the image just 20 times
a second, some people will see TRIPLE-imaging(!) and others will just
see a single object moving jerkily.

Personally, my threshold is somewhere between 15 and 20Hz.  I  can reliably
see triple-imaging (and cinemas' double-image) - but I can't see quadruple
imaging at 15Hz no matter how hard I try.

If you run the CRT at 60Hz and repaint at 10Hz, hardly anyone sees SIX
images moving across the screen - but I have heard of people who see
exactly that.

The actual Hertz rate when your perception switches depends on a lot of
factors - ambient light levels, scene contrast, how tired you are...etc.
But variation between individuals is a significant factor too.

Hence, one person might see cinema double-imaging whilst another merely
sees a jerky/blurry image - and you might find that some movies look
better than others.  If the image is poorly focussed then that would
have an effect too.

-----------------------------------------------------------------------
The second law of Frisbee throwing states: "Never precede any maneuver
by a comment more predictive than "Watch this!"...it turns out that
this also applies to writing Fragment Shaders.
-----------------------------------------------------------------------
Steve Baker                      (817)619-2657 (Vox/Vox-Mail)
L3Com/Link Simulation & Training (817)619-2466 (Fax)
Work: sj...@li...           http://www.link.com
Home: sjb...@ai...       http://www.sjbaker.org

Re: [Algorithms] vrefresh jumpiness

[Madoc E. <tm...@ti...>]

Sounds like you're on to something there. But these wavy lines and double
diagonals don't sound like something you might be able to do in practice,
especially given that the perception doesn't follow constants. You'd have=
 to
get every user to recalibrate with white dots on black backgrounds in dar=
k
rooms every time the framerate or their mood changes.

The only solution I can see is to set a low refresh rate and do your best=
 to
stay above it. We all agree here (at the office) that 60hz graphics with =
the
monitor refresh set higher still looks dreadful.

The other solution is really stiff and jerky control systems, you really
don't notice it much then.

So, crap (excuse my french) graphics that run really fast or crap (pard=F3=
n)
control systems that look jerky anyway. They both seem to be pretty commo=
n
solutions. Next time you see them, think twice before criticising, they
might be the result of extreme cunning.

Cheers,
Madoc

----- Original Message -----
From: "Stephen J Baker" <sj...@li...>
To: <gda...@li...>
Sent: Tuesday, March 30, 2004 7:22 PM
Subject: Re: [Algorithms] vrefresh jumpiness


> Thatcher Ulrich wrote:
> > On Mar 25, 2004 at 05:07 +0100, Nils Pipenbrinck wrote:
> >
> >>If you go into a cinema and watch a movie with 24 or 25 fps you don't
> >>see the ghost images.
> >
> >
> > You should, movies are unbelievably bad in this respect.  Pay
> > attention during any panning shot.  Movies have half-frame-rate-itis
> > all the time, since they always flash the same frame twice.
>
> You are both right.  Almost certainly, one of you sees one thing, the
> other sees something different.
>
> > The best explanation of the double-image artifact I've heard is that
> > your eye tries to smoothly track moving objects.  In the case of 30fp=
s
> > update on a 60Hz screen, with an object moving at a velocity v, your
> > eye tries to track the object at velocity v.
>
> The better way to state this is to imagine the canonical caveman chucki=
ng
> a rock at a rabbit that he fancies for lunch.  As the rabbit runs behin=
d
> a tree, he loses sight of it momentarily - and if his brain didn't deal
> with that, he'd be unable to hit the darned thing.  So after enough
> generations of hungry cavement, we've evolved the ability to interpolat=
e
> the missing data when the object we're tracking vanishes momentarily - =
and
> this is what makes a series of still images in movies and TV appear to =
be
> moving.
>
> I suspect that the lack of interest of some animals in watching TV rela=
tes
> to their failure to have evolved that mechanism.  For them, TV is just =
a
> slide-show.
>
> OK - so what are the implications of this?
>
> If you plot a graph of position against time of an object being updated
> 60 times per second on a 60Hz CRT, the points lie on a nice straight,
> diagonal line - and our brains have no trouble interpolating the
> positions using the million-year-old rabbit-went-behind-a-tree mechanis=
m.
>
> But if you only update the position 30 times a second - but refresh the
> screen at 60Hz and plot a time-versus-position graph, it looks like a
> staircase - right?
>
> There are two ways to draw a conclusion about the position of the objec=
t
> between frames from that data.
>
> 1) Fit a wavy line or an actual square-edged staircase to those points.
>
> 2) Draw two nice straight diagonal lines - one that goes through
>     the odd numbered points - and another parallel line that goes
>     through the even numbered points.
>
> 99% of humans unconsciously do the latter because rabbits can't change
> their velocity at 30Hz.  We just havn't evolved to interpolate that
> kind of motion data.  So we HAVE to take the two-straight-lines view of
> the world and instead of seeing one object moving jerkily - we see TWO
> objects moving along perfectly smoothly right next to each other.
>
> Hence the double-imaging.
>
> However as you drop the update rate, our brains 'snap' to seeing
> the wavy-line/jerky motion version of events because the motion fits
> better to our mental model of a running prey animal.  This seems to hap=
pen
> somewhere below 30Hz and above 10Hz (for 99% of people - there are alwa=
ys
> a few outliers on that bell-curve).
>
> Cinemas are running somewhere in the middle of that range. 24Hz.
>
> That's why *SOME* people see double-imaging in cinemas - and some peopl=
e
> see jerky/blurry motion instead.
>
> On a CRT, if you run the CRT at 60Hz and repaint the image just 20 time=
s
> a second, some people will see TRIPLE-imaging(!) and others will just
> see a single object moving jerkily.
>
> Personally, my threshold is somewhere between 15 and 20Hz.  I  can
reliably
> see triple-imaging (and cinemas' double-image) - but I can't see quadru=
ple
> imaging at 15Hz no matter how hard I try.
>
> If you run the CRT at 60Hz and repaint at 10Hz, hardly anyone sees SIX
> images moving across the screen - but I have heard of people who see
> exactly that.
>
> The actual Hertz rate when your perception switches depends on a lot of
> factors - ambient light levels, scene contrast, how tired you are...etc=
=2E
> But variation between individuals is a significant factor too.
>
> Hence, one person might see cinema double-imaging whilst another merely
> sees a jerky/blurry image - and you might find that some movies look
> better than others.  If the image is poorly focussed then that would
> have an effect too.
>
> -----------------------------------------------------------------------
> The second law of Frisbee throwing states: "Never precede any maneuver
> by a comment more predictive than "Watch this!"...it turns out that
> this also applies to writing Fragment Shaders.
> -----------------------------------------------------------------------
> Steve Baker                      (817)619-2657 (Vox/Vox-Mail)
> L3Com/Link Simulation & Training (817)619-2466 (Fax)
> Work: sj...@li...           http://www.link.com
> Home: sjb...@ai...       http://www.sjbaker.org
>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by: IBM Linux Tutorials
> Free Linux tutorial presented by Daniel Robbins, President and CEO of
> GenToo technologies. Learn everything from fundamentals to system
> administration.http://ads.osdn.com/?ad_id=3D1470&alloc_id=3D3638&op=3Dc=
lick
> _______________________________________________
> GDAlgorithms-list mailing list
> GDA...@li...
> https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
> Archives:
> http://sourceforge.net/mailarchive/forum.php?forum_id=3D6188