Image
From Sunflow Wiki
Contents |
Image Block
There image block dictates the resolution of the image as well as a few other image options. The samples and jitter lines are optional and if they are left out samples will equal 1, contrast will be 0.1, and jitter will be false will be used as the default.
image {
resolution 800 600
aa 0 2
samples 4
contrast 0.1
filter gaussian
jitter false
}
Anti-Aliasing
The aa line is the settings for the adaptive anti-aliasing. These control the under/over-sampling of the image. The image will be first sampled at the rate prescribed by minimum aa value (the first value). Then, based on color and surface normal differences, the image will be refined up to the rate prescribed by the maximum aa value (the second value). Taken from the Sunflow ReadMe file:
A value of 0 corresponds to 1 sample per pixel. A value of -1 corresponds to 1 sample every 2 pixels (1 per 2x2 block) A value of -2 corresponds to 1 sample every 4 pixels (1 per 4x4 block) A value of -3 corresponds to 1 sample every 8 pixels (1 per 8x8 block) ... A value of 1 corresponds to 4 samples per pixel (2x2 subpixel grid) A value of 2 corresponds to 16 samples per pixel (4x4 subpixel grid) A value of 3 corresponds to 64 samples per pixel (8x8 subpixel grid) ...
Examples:
- quick undersampled preview: -2 0 - preview with some edge refinement: 0 1 - final rendering: 1 2
You can turn adaptive anti-aliasing off by setting the min and max values to the same number. For example, an aa of 0 0 will be 1 sample per pixel with no subpixels.
You can You can see a video on explaining what adaptive anti-aliasing is here (.zip).
Samples
Samples are the number of samples. Surprised? When used they indirectly affect many aspects of the scene but directly affects DoF and camera/object motion blur.
Contrast Threshold
There is a line in the image block in which you can change the default contrast threshold. This affects the point at which the renderer decides to adaptively refine between four pixels when doing adaptive anti-aliasing. This line isn't required and the default is usually the right setting. I personally don't see a change in the render with different values.
Filters
If you are oversampling the image (i.e., having the min and max aa positive numbers) you will likely want to use more advanced filters to control the look of the image. The available filters are:
box (filter size = 1) triangle (filter size = 2) gaussian (filter size = 3) mitchell (filter size = 4) catmull-rom (filter size = 4) blackman-harris (filter size = 4) sinc (filter size = 4) lanczos (filter size = 4) bspline (in the 0.07.3 SVN) (filter size = 4)
Check this page out for a good overview of filters. Triangle and box are better for previews since they are faster. The other filters are recommended for final image rendering (my personal favorite is mitchell).
 box |
 triangle |
 gaussian |
 mitchell |
 catmull-rom |
 blackman-harris |
 sinc |
 lanczos |
 bspline |
Jitter
The jitter line (either true or false) turns jitter on or off. Jitter moves the rays randomly a small amount to reduce aliasing that might still be present even when anti-aliasing is turned on. So jittering these pixels makes the aliasing issues become less perceptible. Jitter should be turned off (false) when doing animations because the randomness of jitter makes it very obvious it's on when moving from frame to frame.
Bucket Size/Order
You change the bucket order in the scene file just like you can in the command line. In the scene file, if you want to change the order from the default hilbert (for which you don't need to add any line) you would type the following as it's own line in the .sc file:
bucket 64 column
If you want a reverse order it would look like this:
bucket 48 "reverse spiral"
A larger bucket size means more RAM usage and less time rendering. Usually, a bucket size 64 is a good default - especially if you are using a wide pixel filter like gaussian or mitchell. There are six bucket order types available: hilbert (default), spiral, column, row, diagonal, and random. You can also use these ordered in reverse by adding "reverse" in front of the name. To use reverse, you'll need to use quotes around the reverse order so that the bucket order is still parsed as one token. The number in the line is the pixel size of the each bucket. There is some clamping done internally to make sure the bucket size isn't too small or too big.
Alpha Channels
The SVN 0.07.3 version of Sunflow create alpha channels when exporting to png or tga. Alphas will be recognized in textures in future releases.
Banding
In some cases, and rendered image with a somewhat uniform color will appear to have a banding affect and not be a smooth gradient. The issue is due to the 8 bits per channel used in the rendered image causing "mach banding" (an effect of the human eye) that happens just when values jump from one integer to the next after staying constant for several pixels. In this case the only solution found so far is to use dithering (which only seems to be found in Photoshop) when converting to 24 bit RGB.
Tone Mapping
If you find yourself looking to tone map or you intend to tone map your image before it's even rendered, you may want to consider this tip. Render a bigger image (at least 2x in each direction), then you can reduce the number of samples since the resize (to your originally intended size) will blur the noise. This will avoid having the problem of the high intensity pixels filtering too far out when Sunflow does its filtering. Ideally you would render with aa samples set to 0 0 (fixed at 1 sample per pixel) and a really big image. Then do your tonemapping on that big image and then you can resize to final resolution. The OpenEXR output driver is probably the better choice (vs .hdr) if you want to do this since it can handle arbitrary image sizes (it writes a bucket at a time).
