Illumination Estimation Challenges

Photographer's opinion on the first validation submissions


Urban night scenes are now more significant in photography than before, for two reasons. One is that lighting itself, and signage, has become stronger, more varied and more colourful over the years. The other is that improvements in phone cameras have made it possible for many more people to capture images without effort or planning.

Unlike daytime scenes, however, there has been little or no evolution of perceptual experience as to how such scenes should look in a photograph — for instance, how dark overall, how colourful, how contrasty — and there are no generally accepted standards. By contrast, human vision has evolved under daylight, and we have an innate sense of correctness for colour balance, average brightness, contrast and so on.

The principal characteristics are:
  1. Large unlit, ie dark, areas
  2. Several-to-many point light sources and speculars
  3. Coloured illuminants, including some with a restricted spectrum (such as sodium lamps, eg building facade IMG_7800)
  4. Localised high contrast from light pooling eg building floodlighting
  5. Lighting may be dominated by a single-hue illuminant, or there may be dual-hue illuminants


A. Image integrity

This has mainly to do with artefacting, and could be considered close to being objective, and I propose should be prioritised in evaluation over interpretation:

B. Interpretation

As a professional photographer, I would be looking first to optimise the image according to general industry standards, based on the following criteria:- Memory colours (canonical colours) can help as references. In descending order of usefulness and reliability, for the night scenes here, they are: In addition, I acknowledge that what qualifies as pleasing depends on individual and peer-group taste, and this is flexible. This particularly affects overall colourfulness (which I take to include overall colour cast, the presence of opposing and tangential colours in a scene, and the amount of saturation of these colours), as there are no generally agreed standards as to how a night-time scene SHOULD look. I should stress that this is certainly a concern of professional night-time photography, which at a higher level aims to deliver more than just a standardised record of a scene. It aims to deliver imagery that appeals to viewers. Thus it would be perfectly legitimate from this point of view to render a night-time scene with, for example, a moderate colour cast. This would be an aesthetic decision of the individual photographer.


  1. Artefact-free.
  2. Overall fairly neutral colour balance with colourful small elements. If there is any colour cast, blue is more acceptable, while greens (from cyan to yellow-green) are by tradition unacceptable.
  3. Full tonal range, ie black point set.
  4. Unlit and weakly lit areas dark, ie histogram bias to left
  5. No clipping except for point light sources and speculars.
  6. Saturation not to reach 100%, which reads as unrealistic. This is particularly important for night scenes, featuring both light sources and illuminated small areas against an overall dark background, which enhances brightness. Both the Hunt effect (colourfulness increases with luminance) and Helmholtz-Kohlrausch effect (saturation increases with brightness) help exaggerate these.