On 21-11-2017 13:52, Bald Eagle wrote:
> Thomas de Groot <tho### [at] degrootorg> wrote:
> 
>>> I was wondering what your "yield" was in terms of successful placement of random
>>> points inside.  I did a few real quick experiments, and it seemed like I was
>>> getting _very_ sparse coverage.  I did a (Success/TotalTests)*100 calculation,
>>> and got 1.4%.
>>
>> I don't know about yield. Using VRand_In_Obj() that should be 100%,
>> shouldn't it? Anyway, I just increase or decrease the number of objects
>> and judge the result.
> 
> Yes, I was just testing a vector in the bounding box.
> IIRC VRand_In_Object() is part of an iclude file, so it just does pretty much
> the same thing - the success/fail process is just hidden.

I seem to remember now that that is the way it works indeed.

> 
>> My main focus now is to get a "better" (?) coverage of the arms and legs
>> compared to the body. They seem to be more sparsely populated. So, my
>> idea was to use an additional test: function {pow(f_boxed(x,y,z),2)}
>> provided by Christian Froeschlin some years ago, and thus concentrate
>> the objects more towards the periphery than towards the centre. However,
>> I am not sure what I am doing and I have difficulty scaling this to the
>> correct proportions of the body. Any suggestions there would help me.
> 
> Right - I noticed the same problem.
> I thought about that and naturally what Stephen suggested is the most obvious
> solution.

In the end, that is probably what I am going to do. My other experiments 
give interesting results but not entirely to my liking, and I prefer to 
concentrate on more interesting things ;-)

> Your function-based approach is something that could work, though I was thinking
> of something more cylindrical / spherical / egg-shaped and centered on the body
> trunk.
> 
> Alternatively, you could use an x-y-z nested loop of smaller testing boxes and
> cycle through a random placement in all of those smaller boxes.
> 
> It would, of course, be VERY nice if someone in-the-know could hammer out some
> sort of octree macro/include file that made some sense to those of us who
> haven't had the time to puzzle all of that out yet.
> 
> Along those lines, you could build an array to store coordinates, "scan" the
> bounding box of the figure with an x-y-z nested loop and inside(object, vector)
> tests, and then use the pseudorandom selection process to pick known inside()
> points.
> The advantage there is that you could save the array to disk, and not have to
> test the same figure every single time you wanted to place objects.  You
> wouldn't even need to load the mesh for the figure either.

I am not going to dive too deep into this as it rapidly becomes unknown 
territory to me, but nonetheless I am grateful for the suggestions.

> 
> So far, my whole weekend has been major SNAFU, so I haven't had mush time to
> fiddle with many of the things I've wanted to.  :(
> 
> Hopefully the rest of the week/weekend won't be FUBAR.
> 
Best wishes for that indeed. I have a couple of minor SNAFUs myself at 
present which need sorted out (well, ideally, they /should/ sort 
themselves out eventually). RL... :-/


-- 
Thomas

Post a reply to this message