< Previous | Contents | Manuals Home | Boris FX | Next >
Floating-Point Images
SynthEyes can handle floating-point images from ARRIRAW, Blackmagic RAW, DNG, DPX, EXR, PSD, and TIFF image formats. Floating-point images offer the greatest accuracy and dynamic range, at the expense of substantially greater memory requirement and processing time. DPX images will offer the highest performance, similar to BMP for 8-bit images.
Floating point images may use 32-bit floats, or the 16-bit “half” format. The half format does not have as much dynamic range, but it is almost always enough for practical work even using High-Dynamic-Range images. Half-floats are half the size, only 16 bits, so you can fit more images into memory simultaneously. On modern processors, it takes only a little extra time to translate between the half format and an 8 bit, 16-bit, or float format you can track or display.
SynthEyes offers separate bit-depth selections for processing and for storage. If you need the extended range of a float (or 16-bit int) format, you can use that for any processing (especially gamma correction and 3-D LUTs), to reduce banding, then select a smaller format, Half, 16-bit, or 8-bit, for storage. Some additional processing time will be required.
Though a floating-point image—float or half—provides accuracy and dynamic range, to track or display it, it must be converted to a standard 8-bit or 16-bit form, albeit temporarily. To understand the necessary controls, here are a few details on how that is done (industry-wide).
Eight and sixteen bit (unsigned) integers are normally considered to range from 0 to 255 or 65535. But to convert back and forth, the numbers are considered to range from 0 to 1.0 (in steps of 1/255), or 0 to 1.0 (in steps of 1/65535).
Correspondingly, the most-used values of the floating-point numbers ranges from 0 to 1.0 also. With all the numbers ranging from 0 to 1, it is easy to convert back and forth.
But, the floating point values do not necessarily have to range solely between 0 and 1. With plenty of dynamic range in the original image, there may be highlights that may be much larger, or details in the shadow that are much lower. The 0 to 1 range is the only portion that will be converted to or from 8- or 16-bit.
The F.-P. Range Adjustment (F.-P. for floating-point) on the Shot setup dialog allows you to convert a larger or smaller range of floating-point numbers into the 0 to 1 range where they can be inter-converted. The effect of this control is to brighten or darken the displayed image, but it affects only the conversion to integer for display and tracking—not the values themselves.
You can adjust the F.P. Range Adjustment, and it will not affect the floating- point images later written back to disk after lens distortion or stabilization.
This is quite different than the Exposure control on the Color tab. The Exposure control changes the actual floating-point values that will be written back to disk later.
The two controls serve different purposes, though the end result may appear the same at first glance.
NOTE: The Kurves color correction has another more advanced way to deal with larger floating point values. You can adjust the range control knobs and set your splines to map the maximum value (see Auto-set on the right-click menu to help) either down 1.0, or also set the output maximum range to the same value, preserving (with possible modifications) the overrange values.
©2024 Boris FX, Inc. — UNOFFICIAL — Converted from original PDF.