Qtpfsgui Notes - Pattanaik TMO

Following a brief sojourn in Scotland, I append my notes on the Pattanaik TMO. Comments and crits are welcome as usual.

Qtpfsgui Pattanaik Notes

Test Image: Test1.hdr (Hass Burn images 357, 359, 361, April 2009 copyright DARW)

Initial Set-Up: As per Drago Notes, selecting Pattanaik from the drop-down menu for TMO's.

Pattanaik Tone Mapping Operator

This is one of the more complex TMO's in that there are three slider parameters and two radio button controls as well as the option to alter pre-gamma and use the Adjust Levels tool. The authors are S N Pattanaik, J Tumblin, H Yee, and D P Greenberg. However, from the sources listed under the Drago post there is very little information on how the user is to operate this algorithm. One can infer that it is meant to mimic the way the human eye responds to colour and light/dark.

The slider parameters are:

Multiplier, default 1.000, range 0.001 – 50.000;
Cone Level, default 0.50, range 0.00 – 1.00; and
Rod Level, default 0.50, range 0.00 – 1.00.

The radio box controls are:

Local Tone Mapping and
Cone and Rod based on Luminance.

Activating either of these buttons, greys out the other button and the Cone Level and Rod Level sliders.

Starting with the default parameters, the Pattanaik TMO yields a mostly white blown image with only the darkest portions of the HDR showing anything reasonable. Lowering the pre-gamma to any other value seems to make matters worse, but the image starts to emerge at higher pre-gamma values of 1.5 - 2.0 or more. With pre-gamma back at default, setting the Multiplier to its minimum (0.001) gives a totally desaturated image; colour starts to be restored with this parameter set to about 0.100, but even so much of the image is washed out to white. Adjustment of Multiplier to 0.100 and pre-gamma to 1.500 gives a desaturated but recognisable image.

Activating the Local Tone Mapping option, with Multiplier at default, gives a very dark but coloured image. Using the Adjust Levels tool provides a usable image, but one that is clearly non-realistic with overly contrasted areas. Increasing the Multiplier parameter to 7 in this instance gives a modestly good image, but lacking saturation and obviously a tone mapped HDR image. Thus, so far, the Local Tone Mapping option gives useful results with the Multiplier set substantially above its default.

Returning to default values and now activating the Cone and Rod based on Luminance option leads to a dark, but promising image. However, altering the other possible parameters (Multiplier and pre-gamma) did not lead to anything better in my hands.

From the defaults altering the Cone Level slider up or down appears to make no significant difference to the generally blown image. Similar results were found from altering the Rod Level slider. However, recalling that higher pre-gamma settings were useful, setting pre-gamma to 2.50 and the Rod and Cone Levels to 0.90 gives a somewhat desaturated image, but with reasonable tonal variation and detail. Use of the Adjust Levels tool helps the tonal range.

This TMO appears to be very sensitive to the HDR file itself, in that repeating the above tests on a different .hdr file gave different results. This lack of consistent response is not helpful.


Conclusions:

The lack of saturation and complex set of parameters to adjust make this TMO one to be avoided, but images with good tonal range can be produced either with Rod and Cone Levels set to 0.9 or more along with a high pre-gamma, or by using the Local Tone Mapping option.

Technical Points:

There are several papers by Pattanaik and his co-workers, see web address in Sources below. The key work appears to be S. N. Pattanaik, Jack E. Tumblin, Hector Yee, Donald P. Greenberg, "Time-Dependent Visual Adaptation for Realistic Real-Time Image Display",**Proceedings of SIGGRAPH 2000, pp. 47-54, New Orleans, 23-28 July, 2000.

Unfortunately, while there are references to rods, cones, and so on, the paper does not appear to address the parameters of the TMO as presented to the user. There are, however, a couple of points of interest. First, the paper assumes that CRT display devices are being used (publication date 2000, TFT's were not really available). Second, there is reference to the display device (i.e. CRT) having a gamma of 1.0. Third, there are assumptions made about the visual response of rods and cones to intensities of light. All of these points suggest to me that with modern TFT monitors and with personal variations of visual response, the Pattanaik TMO may need updating.

Sources:

http://www.cs.ucf.edu/~sumant/publications.html