This is a really interesting graphic, I did not know cameras mostly consisted of lenses. Now I am wondering how the camera computational interprets its given input. I wonder what its OS/Embedded System looks like. I heard that modern cameras use AI to make the images better. I am wondering how exactly that works. Hopefully CS180 will be offered again sometime in the future, as computational photography is very interesting to me. I have been told several times that graphics is essentially solved, but I suppose its not entirely solved yet in the context of AI.
snowshoes7
I agree. I didn't realize that most of what's happening inside a camera is lenses refracting light so that it can be picked up by a sensor--and what Wil said above makes me think, if AI is being used in the context of image processing in the camera itself, doesn't that mean that information is being lost? Not just in the fact that an image sensor has a limited size and can only pick up so much information of the image it receives, but then in this post-processing step?
S-Muddana
I think it is really cool how we can fit millions of these tiny light-sensitive pixels that convert light coming in to electrical signals. Having all those in a professional camera is one thing, but fitting them in a smart phone is amazing.
stang085
I never knew how many parts went into making a camera, it's really interesting to see how they all come together to get you different styles and images
rcorona
To respond to @snowshoes7's comment above, I like to think that not all post-processing operations entail a loss of information, but rather a transformation or in some cases a recovery of it.
For example, I'm reminded of tasks such as image denoising (https://arxiv.org/pdf/2006.11239.pdf) or colorization (https://arxiv.org/abs/1603.08511), where a system is tasked with "adding back" information into an image signal. Although the original signal might be noisy, the model can learn priors from data which guide how it fills in or post-processes the image (e.g. prior experience with dogs can inform us on how a noisy image of a new dog probably looks like).
yangbright-2001
Cameras offer many different modes to choose when we are using them, such as special modes for taking photos of a person, or taking photos of objects very close to the camera. I am curious about how these different modes and styles settings are achieved. Are they done by "AI" after the photo was taken, or by the lens themselves?...
jonnypei
The various different modes in the camera are partly by adjusting the camera parts, and partly via AI (for example for super-resolution). And a lot of the other shading/picture editting, there's a lot of in-device software involved.
Alescontrela
This is a fascinating image, I didn't know how complicated the internals of a camera were, nor just how many lenses are contained within.
This is a really interesting graphic, I did not know cameras mostly consisted of lenses. Now I am wondering how the camera computational interprets its given input. I wonder what its OS/Embedded System looks like. I heard that modern cameras use AI to make the images better. I am wondering how exactly that works. Hopefully CS180 will be offered again sometime in the future, as computational photography is very interesting to me. I have been told several times that graphics is essentially solved, but I suppose its not entirely solved yet in the context of AI.
I agree. I didn't realize that most of what's happening inside a camera is lenses refracting light so that it can be picked up by a sensor--and what Wil said above makes me think, if AI is being used in the context of image processing in the camera itself, doesn't that mean that information is being lost? Not just in the fact that an image sensor has a limited size and can only pick up so much information of the image it receives, but then in this post-processing step?
I think it is really cool how we can fit millions of these tiny light-sensitive pixels that convert light coming in to electrical signals. Having all those in a professional camera is one thing, but fitting them in a smart phone is amazing.
I never knew how many parts went into making a camera, it's really interesting to see how they all come together to get you different styles and images
To respond to @snowshoes7's comment above, I like to think that not all post-processing operations entail a loss of information, but rather a transformation or in some cases a recovery of it.
For example, I'm reminded of tasks such as image denoising (https://arxiv.org/pdf/2006.11239.pdf) or colorization (https://arxiv.org/abs/1603.08511), where a system is tasked with "adding back" information into an image signal. Although the original signal might be noisy, the model can learn priors from data which guide how it fills in or post-processes the image (e.g. prior experience with dogs can inform us on how a noisy image of a new dog probably looks like).
Cameras offer many different modes to choose when we are using them, such as special modes for taking photos of a person, or taking photos of objects very close to the camera. I am curious about how these different modes and styles settings are achieved. Are they done by "AI" after the photo was taken, or by the lens themselves?...
The various different modes in the camera are partly by adjusting the camera parts, and partly via AI (for example for super-resolution). And a lot of the other shading/picture editting, there's a lot of in-device software involved.
This is a fascinating image, I didn't know how complicated the internals of a camera were, nor just how many lenses are contained within.