Personal View site logo
MIT HDR camera
  • 12 Replies sorted by
  • These "simulated" conventional camera exposures are hilariously inaccurate. They are trying way too hard to make their product look better.

  • May be, but idea is simple.

  • I find it quite interesting, counting the buckets can lead to infinite dynamic range. It could be the future.

  • so logical. :) great stuff. if it can be implemented for a small amount of money in some years. word @jazzroy .

  • @joethepro: If you understand the concept, you'll see they have a point and see how their idea is ridiculous simple, yet brilliant. http://web.media.mit.edu/~hangzhao/papers/moduloUHDR.pdf

    Besides, simulating the concept is utterly trivial, you just take some conventionally captured image and truncate it to it's N least significant bits. The hard part was deriving the unwrapping algorithm, but I've also seen harder math. Proving the concept is really simple once you knocked the unwrapping, yet ...

    ... what I'm really curious about is how extensively they've run their algo on long exposure noisy images as I can't see a trivial solution on their concept modulus camera on how to perform dark frame subtraction. Noisy images won't provide the smooth gradients their energy minimization algo strives on, so it might be either too much of a challenge, compromise or limitation.

  • @duartix What are you talking about? I know this, just saying they took it too far, but I guess it was to prove a point to those who dont understand all this.

  • The innovation here is not an image sensor that resets each bucket when it becomes full. So-called modulo sensors already exist. The innovation is an algorithm that constructs a high dynamic range image from the modulo read-out of each pixel without knowing how many times each bucket was reset.

    BTW, a modulo sensor is fundamentally different from a typical camera sensor. A typical sensor collects charge over the exposure period, and then measures the charge all at once. A modulo sensor continuously collects charge over the exposure period, but dumps it in multiple small increments throughout the exposure, counting the increments. The count is the radiance measurement. The counter's finite number of bits is the modulo nature of the output. I don't know how such a sensor compares to a standard camera sensor, practically. I believe they are just experimental.

  • @balazer

    Yep, extreme version of such sensors are binary sensors that measure presence of charge (above some threshold) in pixel over very short time period (1- present, 0- absent) and output this value. You can have huge DR with such architecture (as it is restricted only by noise floor from bottom and catched electrons count during exposition from top).

  • @balazer: I think they know how often they read out and "empty the buckets". why wouldnt they?

  • There is limited memory at each photo site. Just like a car's odometer, at some point the counter rolls over to 0.

  • I dont understand or care about the science but I imagine in time cameras will have a dynamic range setting you can just dial in how much you want. All good.

  • "I imagine in time cameras will have a dynamic range setting you can just dial in how much you want."

    I believe this will happen. The Nikon honeycomb sensor specs says it can combine the subpixels to increase DR from 1 to 3 or 5 fstops (without changing iso ou shutter speed) maybe more changing iso... So they can implement a dial button or a menu item to change how many fstops we want, something like Sony DRO - dynamic range optimizer, but much more powerful. Other sensor designs can do similar things I think...

    Increase DR and keep blacks to be black and to keep same perceiving gamma, without post production. Similar thing as I did in the Cook Picture Styles for Canon which works similar to sony DRO.