February 07, 2022

Present

• Alexander Schwank
• Michael Johnson
• Nick Porcino
• Kimball Thurston
• Quentin Birrer
• Sergio Rojas
• Laura Scholl
• David Aguilar

Discussion

• PhysLight (Kimball Thurston, Weta)
  • PhysLight github repo from Siggraph 2020: https://github.com/wetadigital/physlight
  • PhysLight document now has a substantially complete physical characterization of the spectral transport of photons from emitter to sensor.
  • Weta has built some calibration gadgets, such as the so-called "lightsaber" to generate data characterizing elements of their production set up
  • It's preferable that the PhysLight model be introduced to Usd at a foundational level, not at an optional schema level
  • The "terabytes of data" Netflix has need some documentation around them about what the data is, how to index it, which data has longevity, which data has usefulness across more than one camera
  • The transition to pbr workflows took years for artists to be comfortable with painting albedo instead of "textures". The new Physlight work needs to sit along side unitless rendering workflows as a transition is likely to take years.
  • MaterialX 1.38 is still unitless, even the EDF is intended as a scale on a unitless light. Coordination between a next version of UsdLux, MaterialX, and Physlight is a conversation that needs to happen.(see https://github.com/materialx/MaterialX/blob/main/documents/Specification/MaterialX.v1.38.PBRSpec.pdf )
  • There is no common practice on metrology of cameras or lenses. In order for a scheme like Physlight to be viable, measurements should be reproducible. It may be necessary to design new metrology around focal length and clear definition of nominal points such as what exactly does "sensor plane" mean? Is it the air/crystalline interface? is the crystalline/sensor site interface? Is it the top of the microlenses affixed over the sensor, and so on. How do we measure from the calibration mark on a camera body to a locational on the stage? How about from the mark to the lens flange? To the sensor plane? To the IMU?
  • The relationship of Physlight to existing renderers is an open question. If we open a Physlight document in a DCC, is there anything we can do so that the first light in the viewport is not completely blown out?
  • The sensor area is well defined in Physlight, the sensor sites are not. Bayering (and sensor site layout in general) is a problem in specifying the sensor site. Given the reluctance of manufactures to share or commit specifically to sensor site layout, is there some measured value that might in fact be preferable, such as characterization of the Airy disc or diffraction limits?
• Next steps
  
  • Kimball to update the PhysLight document and share with this group before next meeting in 4 weeks