4 ms·

Maybe it would help see the more important difference if they showed what video looks like without any such transformation, or the typical post-processing someo

Maybe it would help see the more important difference if they showed what video looks like without any such transformation, or the typical post-processing someone would do without this tool? I.e., I'm not comparing against film, I'm comparing against what comes out of the video camera.

wilg 6y ago
This type of comparison is a pretty natural one to want. We are still thinking about ways to best demonstrate Filmbox. The best way is to use it, but we will try to come up with a way to show this. This particular comparison raises some interesting philosophical questions, which is why we haven't gotten to it yet. The comparison could be pretty misleading if done wrong. Filmbox is designed to produce an accurate film look from scene-referred footage. But digital image data really has no look in any meaningful way. We could, for example, show the video in a log color space as it’s encoded, but that's an arbitrary encoding that is not even intended for display. We could apply some "video" LUT or simulate how someone might "typically" color grade the footage, but that's a creative choice - and one that can still be performed in addition to the Filmbox emulation pipeline. The right way to think about it is perhaps that video can be prepared to look like anything, But modern motion picture film has a fairly defined look. Filmbox is designed to provide ways of processing video that are closest to processed film. So we feel the meatiest comparison is Filmbox to actual film.
- supernova87a 6y ago
  That is a good point, I guess any output has some transformation applied, just a question of what you are aiming for. Maybe the best way to put it could be, "if someone tried to get it to look as much as film as they could, what would they lack that your tool provides"?
  - wilg 6y ago
    The answer to that probably gets a bit tautological and sounds cheeky – if they did a great job making it look like film, then nothing! But in practice without a clear target and a lot of empirical data about the various properties of photochemical imaging they would end up with a subjective look that may look filmic on a limited range of shots but would not represent a dynamically functional model of the response of the photochemical process. This is why we think the most apt comparison is our output compared with film, because that's the target look. We don't think the existing solutions do as good of a job of it as easily as Filmbox.

wilg 6y ago

This type of comparison is a pretty natural one to want. We are still thinking about ways to best demonstrate Filmbox. The best way is to use it, but we will try to come up with a way to show this. This particular comparison raises some interesting philosophical questions, which is why we haven't gotten to it yet. The comparison could be pretty misleading if done wrong. Filmbox is designed to produce an accurate film look from scene-referred footage. But digital image data really has no look in any meaningful way. We could, for example, show the video in a log color space as it’s encoded, but that's an arbitrary encoding that is not even intended for display. We could apply some "video" LUT or simulate how someone might "typically" color grade the footage, but that's a creative choice - and one that can still be performed in addition to the Filmbox emulation pipeline. The right way to think about it is perhaps that video can be prepared to look like anything, But modern motion picture film has a fairly defined look. Filmbox is designed to provide ways of processing video that are closest to processed film. So we feel the meatiest comparison is Filmbox to actual film.

supernova87a 6y ago
That is a good point, I guess any output has some transformation applied, just a question of what you are aiming for. Maybe the best way to put it could be, "if someone tried to get it to look as much as film as they could, what would they lack that your tool provides"?
- wilg 6y ago
  The answer to that probably gets a bit tautological and sounds cheeky – if they did a great job making it look like film, then nothing! But in practice without a clear target and a lot of empirical data about the various properties of photochemical imaging they would end up with a subjective look that may look filmic on a limited range of shots but would not represent a dynamically functional model of the response of the photochemical process. This is why we think the most apt comparison is our output compared with film, because that's the target look. We don't think the existing solutions do as good of a job of it as easily as Filmbox.