stevenrobertson/cuburn

PyCUDA implementation of a GPU-accelerated fractal flame renderer.

cuburn's Issues

FLAM3 Background color is not respected

Lil' sickos like me, sometimes like to rape the beauty of flames by creating black ones, but with no other background color possible then black(with jpegs), this gets a bit too black, even for my taste.

Rendering transparent .png's and hacking some shit together with imagemagick is not an option on so many renderings and for some other reasons, so a fix would be welcome.

Still being supported?

Is cuburn still being supported by anyone? I don't see any issues posted in almost 3 years. I was wondering if there are plans to migrate it to Python 3 (it looks like it was written for Python 2, although no mention is made in the requirements).

Thanks.

None of the video output formats work

All Debian derivates and with that most of the popular linux distros, sadly got infected with avconv and stabbed ffmpeg in the back, so nothing works anymore and you maybe should change the scripts to avconv, since breaking installation discipline to get CUBURN outside of the main repository is already enough sin for one life :D

Divide by zero in convert.py

As i already wrote you in a mail almost a month ago, most of my renderings(with estimator_radius="0") were ignored because of a divide by zero in genome/convert.py @ somewhere around line 147 after the division of course, what i somehow managed to cheap hack patch with
('filters.de.minimum', lambda d: (float(d['estimator_minimum']) / (float(d.get('estimator_radius', 11))+0.001)) if 'estimator_minimum' in d else None),
but since i have a severe python phobia, it would be nice if you could implement a proper divide by zero protection.

Missing dependencies

This dependency is not mentioned in README.md:

pyglet

Cant render .flame files

I have files ( edges ? ) which end with ".flame" and start with < sequence> tag. cuburn cant render them

Issues with CUDA

To be able to use CUBURN I had to go trough a damn hell ride to get everything "working" with a nvidia "optimus" "powered" gaming laptop on a MINT17 linux, but any CUDA app lags my whole desktop so much, that the cursor only moves once every second, that is of course not your problem, but since we on linux don't like that NVIDIA binary blob crap too much, i just wanted to ask, if you maybe know, if it would be possible these days, to port CUBURN to the new GPGPU Compute-Shader in the new OpenGL spec or at least somehow to GLSL?(If you even would do it, because it would make CUBURN more hardware/software independend, then please do it in C/C++, shitty python is the only reason i can't commit something to CUBURN)
I mean, these freaks do realtime and even interactive mandelbulb spacetown journeys in GLSL, which don't fuck my system that much and it even runs in a lousy browser without having to install anything besides the normal graphics drivers(What mostly works relatively well compared to CUDA/OpenCL)
http://glslsandbox.com/e#28950.0

KeyError: 'chaos'

Did a test on 1063 FLAM3 files with the following results:
19 of them were completely ignored because CUBURN obviously didn't had implemented the very important flam3 CHAOS param as we can see here->
§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
Traceback (most recent call last):
File "main.py", line 188, in
main(args, prof)
File "main.py", line 100, in main
gnm, basename = gdb.get_anim(args.flame, args.half)
File "cuburn/genome/db.py", line 49, in get_anim
gnm = convert.node_to_anim(self, gnm, half=half)
File "cuburn/genome/blend.py", line 17, in node_to_anim
node = resolve(gdb, node)
File "cuburn/genome/blend.py", line 50, in resolve
sp = resolve_spec(spec, k.split('.'))
File "cuburn/genome/util.py", line 72, in resolve_spec
sp = sp[name]
KeyError: 'chaos'
§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§§
Another 49 just rendered a fat pink dot, further investigations i did on these, led to no results till now, no error messages, not a matter of how many formulas in variation, not the amount of Xforms, so i am a bit stuck there.
And there are also many of them which are not rendered as they should, but those i didn't counted yet.
Can it be, that CUBURN also does not support all the variation formulas of the FLAM3 spec, if so, which one not?
The complete list of variations i get listed->
0.linear
1.sinusoidal
2.spherical
3.swirl
4.horseshoe
5.polar
6.handkerchief
7.heart
8.disc
9.spiral
10.hyperbolic
11.diamond
12.ex
13.julia
14.bent
15.waves
16.fisheye
17.popcorn
18.exponential
19.power
20.cosine
21.rings
22.fan
23.blob
24.pdj
25.fan2
26.rings2
27.eyefish
28.bubble
29.cylinder
30.perspective
31.noise
32.julian
33.juliascope
34.blur
35.gaussian_blur
36.radial_blur
37.pie
38.ngon
39.curl
40.rectangles
41.arch
42.tangent
43.square
44.rays
45.blade
46.secant2
47.twintrian
48.cross
49.disc2
50.super_shape
51.flower
52.conic
53.parabola
54.bent2
55.bipolar
56.boarders
57.butterfly
58.cell
59.cpow
60.curve
61.edisc
62.elliptic
63.escher
64.foci
65.lazysusan
66.loonie
67.pre_blur
68.modulus
69.oscilloscope
70.polar2
71.popcorn2
72.scry
73.separation
74.split
75.splits
76.stripes
77.wedge
78.wedge_julia
79.wedge_sph
80.whorl
81.waves2
82.exp
83.log
84.sin
85.cos
86.tan
87.sec
88.csc
89.cot
90.sinh
91.cosh
92.tanh
93.sech
94.csch
95.coth
96.auger
97.flux

Cuburn always appends _0002.png on still images

Another minor problem is, if i want to render a still frame/image, and specify a name for that .png file like so:
python main.py "$0" --resume --still --codec png --width 1280 --height 720 -n "/tmp/${outname}.png"
CUBURN always append _0002.png no matter what i say, what completely defies the sense of being able to have a set name param.

ValueError: could not broadcast input array from shape (179) into shape (16384)

Hi

I'm trying to run cuburn to render some flames but I'm getting the next error when execute it:

>python main.py test.flame
Traceback (most recent call last):
  File "main.py", line 180, in <module>
    main(args, prof)
  File "main.py", line 122, in main
    rmgr = render.RenderManager()
  File "cuburn\render.py", line 245, in __init__
    self.fb = Framebuffers()
  File "cuburn\render.py", line 98, in __init__
    seeds = mwc.make_seeds(util.DEFAULT_RB_SIZE * 256)
  File "cuburn\code\mwc.py", line 41, in make_seeds
    seeds[0][:] = mults[:nthreads]
ValueError: could not broadcast input array from shape (179) into shape (16384)

My system specs are:

• Intel Core 2 Duo E8400
• Nvidia GTX660
• Windows 7 SP1 64-bits

And the dependencies that I've installed to run cuburn:

• Python 2.7.7-win32
• Boost 1.55.0 (uncompressed and added to path)
• pycuda 6.0.37-win32
• numpy 1.8.1-win32
• scipi 0.14.0-win32
• tempita 0.5.2 (installed with pip from PyCharm)
• PIL 1.1.7 (this one wasn't on the requirements but an error ask for it)
• Visual Studio 2008 express (also 2012 and 2012 express until I finally get C compiler to work)
• Nvidia Cuda Toolkit 6.0
• Nvidia Graphic Drivers 337.88 (latest at the date 12 Jul 2014)

I know a little bit about Python and never worked with numpy, scipy or any of the other libraries (cuda, boost...), so I don't know what could be going wrong (as far as I search, the error looks related to numpy).
I'm working on a project that uses flames a lot for data visualization. I found the paper Accelerated rendering of fractal flames and finally ended with cuburn, which could save me a lot of time of rendering several images.

Any help is much appreciated.
Thank you!

EDIT:
I forgot to tell that my input flame is a very basic flame with just one linear transformation withouth any modifications, created with Apophysis 2.09

<Flames name="test">
<flame name="mybasictest" version="Apophysis 2.09" size="784 757" center="0 0" scale="196" oversample="1" filter="0.2" quality="1" background="0 0 0" brightness="4" gamma="4" gamma_threshold="0.04" >
   <xform weight="0.5" color="0" linear="1" coefs="1 0 0 1 0 0" />
   <palette count="256" format="RGB">
      FF0000D31616BD2121A72C2C9137377C4242714747664D4D
      ...
      ...bunch of gradient numbers here omited...
      ...
      9400009000008C00008800008400008000007C0000750000
   </palette>
</flame>
</Flames>

Damn nasty problem with what one can render with CUBURN

There is also a damn nasty problem with what one can render with CUBURN,
so nasty that i think it is its biggest downside and that you really should change it.
The problem is, you can only have a black background on jpeg output, but jpeg looks terrible washed-out compared to png, which sadly only can be rendered with alpha transparency for now, what kills the intensity of the colors by making them transparent, what seems NOT to be possible to change in post processing, so for example, if i open the transparent CUBURN PNG render result of the truly godlike flame from the electric sheep website(see attachment) in Gimp to underlay a black background to merge both layer together, to then finaly be able to kill the alpha channel without having a white background(stupid Gimp!)... but then i don't get that much beautiful colors and shapes as when i had rendered it with CUBURN as jpeg or even better with Qosmic(Flam3) as PNG without transparency("Early Clip" is also very interesting in some cases). So there seems to be a clear difference between directly alphaless rendered png and the once which had alpha whilst rendering, therefore please FIX this. (If you implement this PNG transparency and early clip yes/no param, please make also, that we can choose the color of the background, guess flam3 header tag supports it as far as i can see)

Problem_Test_Flame.flam3.zip