4 ms·

A great misinterpretation of the photography is that it's an objective medium. Any combination of lenses and film stocks (or equivalent) is going to represent b

A great misinterpretation of the photography is that it's an objective medium. Any combination of lenses and film stocks (or equivalent) is going to represent but a flat, skewed representation of the three-dimensional world; it's been interpreted before anyone performs any processing, computational or analog. Susan Sontag's "On Photography" is a great read on this topic for anyone marginally interested in not just photography, but art in general.

LegitShady 6y ago
>Any combination of lenses and film stocks (or equivalent) is going to represent but a flat, skewed representation of the three-dimensional world this is no different than any other sensor. It doesn't mean replacing data with guesses is better than the sensor representation of the world, which is the issue at hand today.
- strogonoff 6y ago
  Interesting thing’s happening here. Previously there was a clear line between scene-referred image data, which was treated as objective record of a 2D slice from a 3D world by way of measuring light, and output-referred image data—one of the countless lossy adaptations of that data to fit the limitations of some particular medium (display, paper, etc.) in order to be actually viewed. The scene- to output-referred data conversion is where objectivity inevitably went out the window, but not earlier—the original scene-referred data was mostly treated as immutable. What these guys are doing actually happens at the demosaicing stage, and from what I understand the resulting “super resolution” image is still scene-referred—but it isn’t representing the actual captured light anymore! In other words, we’ll have raw images that are partially “guesses” and no longer an objective record. This isn’t necessarily good or bad, but is somewhat of a paradigm shift I’d say. As a side note, I wish Adobe released the mechanism so that it could be made one of the demosaicing methods available in open-source raw processors, but I take it this won’t be likely.
  - Rastonbury 6y ago
    Interesting thought, does using a physical fisheye lens that warps in different way than the eye actually sees the same using a computer-generated fisheye effect on a source image? Or any other sort of analog filter/lens that changes the picture for that matter
    - strogonoff 6y ago
      > Does using a physical fisheye lens that warps in different way than the eye actually sees the same using a computer-generated fisheye effect on a source image? I can’t parse this for some reason, could you rephrase? > Or any other sort of analog filter/lens that changes the picture for that matter The glass on the camera affects the shape of the 2D slice of the 3D world captured by the camera and can attenuate light of different frequencies; technically the better you know which equipment was used, the stronger the element of objectivity to raw data captured by camera sensor. Addendum: I take back most of my original comment. Adobe’s super resolution tool operates at the demosaicing stage, so image data it produces is no longer strictly scene-referred (it can’t be both demosaiced and scene-referred). The actual sensor capture is still scene-referred. (I think we’ll see tools that take scene-referred data and output scene-referred data eventually though.)
baq 6y ago
more or less the same thing with sound and audiophiles. it's a lucrative business of selling placebos, dreams and hope, though.

LegitShady 6y ago

>Any combination of lenses and film stocks (or equivalent) is going to represent but a flat, skewed representation of the three-dimensional world this is no different than any other sensor. It doesn't mean replacing data with guesses is better than the sensor representation of the world, which is the issue at hand today.

strogonoff 6y ago
Interesting thing’s happening here. Previously there was a clear line between scene-referred image data, which was treated as objective record of a 2D slice from a 3D world by way of measuring light, and output-referred image data—one of the countless lossy adaptations of that data to fit the limitations of some particular medium (display, paper, etc.) in order to be actually viewed. The scene- to output-referred data conversion is where objectivity inevitably went out the window, but not earlier—the original scene-referred data was mostly treated as immutable. What these guys are doing actually happens at the demosaicing stage, and from what I understand the resulting “super resolution” image is still scene-referred—but it isn’t representing the actual captured light anymore! In other words, we’ll have raw images that are partially “guesses” and no longer an objective record. This isn’t necessarily good or bad, but is somewhat of a paradigm shift I’d say. As a side note, I wish Adobe released the mechanism so that it could be made one of the demosaicing methods available in open-source raw processors, but I take it this won’t be likely.
- Rastonbury 6y ago
  Interesting thought, does using a physical fisheye lens that warps in different way than the eye actually sees the same using a computer-generated fisheye effect on a source image? Or any other sort of analog filter/lens that changes the picture for that matter
  - strogonoff 6y ago
    > Does using a physical fisheye lens that warps in different way than the eye actually sees the same using a computer-generated fisheye effect on a source image? I can’t parse this for some reason, could you rephrase? > Or any other sort of analog filter/lens that changes the picture for that matter The glass on the camera affects the shape of the 2D slice of the 3D world captured by the camera and can attenuate light of different frequencies; technically the better you know which equipment was used, the stronger the element of objectivity to raw data captured by camera sensor. Addendum: I take back most of my original comment. Adobe’s super resolution tool operates at the demosaicing stage, so image data it produces is no longer strictly scene-referred (it can’t be both demosaiced and scene-referred). The actual sensor capture is still scene-referred. (I think we’ll see tools that take scene-referred data and output scene-referred data eventually though.)

baq 6y ago

more or less the same thing with sound and audiophiles. it's a lucrative business of selling placebos, dreams and hope, though.