4 ms·

I bought an old prosumer point and shoot recently to take some more thoughtful pictures. It's fairly limited by the CCD sensor at 12MP. My phone, a GS22, takes

I bought an old prosumer point and shoot recently to take some more thoughtful pictures. It's fairly limited by the CCD sensor at 12MP. My phone, a GS22, takes sharper pictures, automatically does the cool aperture settings for a foreground focus, blurry background, and other fancy things. But when you zoom in, there's a ton of terrible upscaling blur that sorta degrades the overall shot. Additionally it just isn't that comfortable trying to shoot on a phone. I'm in the school of thought that purposeful technology makes you think a lot more about the role of technology. I like using technology, I enjoy learning to use it like a tool. A too-good phone camera is great for reference shots, quick social media shots, etc, but isn't enjoyable at all when your goal is to take pictures for the sake of taking pictures. Over-optimization really strips the joy out of life sometimes.

jebarker 3y ago
Not only do phones have the upscaling blur, there's also a telltale "painterly" sign of convolutional neural network based processing that I see everywhere now especially in Google products.
- MSFT_Edging 3y ago
  It honestly makes some photos lose detail. Something you could make out in pixelated form just gets destroyed in the CNN processing.
  - jebarker 3y ago
    That's an interesting observation that human vision can make out detail from a noisy signal. Non-generative CNNs are typically going to filter noise due to the loss functions used. Generative models could in theory learn to recognize whatever signal in that noise humans are queuing on and guess at the missing content. I think there'd be resistance to generative models in photography pipelines though based on the "made-up pixels" argument.
vel0city 3y ago
> automatically does the cool aperture settings for a foreground focus, blurry background, and other fancy things. Its not actually doing any cool aperture settings, its faking it all with software. And depending on the software stack and the picture in question, it can do a pretty bad job at it. I do appreciate the effects at times and use them myself, but on many shots I'll notice the depth of field effects being used in places where it shouldn't like too far on the shoulders, turning the edges of someone's hair to mess, a secondary subject next to but just slightly behind the main subject getting radically out of the fake depth of field, etc. Most of the photos I take day to day are on my phone's camera, for sure. But when I'm actually wanting to really take some pictures, I bust out an actual camera. My current main goto is my G9x Mark II, which is a lot of fun and can still send photos straight to my phone after shooting so I can easily send them to people right away.
- MSFT_Edging 3y ago
  Yeah I'm just talking about this at a high level. Its basically the same problem, so much auto-processing is done to make pictures generally look good, so that so many intentional pictures just look disappointing.
rob74 3y ago
Not wanting to carry an assortment of lenses with me, I have used bridge cameras for ages, and my current one (https://en.wikipedia.org/wiki/Panasonic_Lumix_DMC-FZ1000 https://en.wikipedia.org/wiki/Panasonic_Lumix_DMC-FZ1000) is pretty well-reviewed and generally does make really good photos, but my 6-year-old Pixel 2 phone still takes better HDR pictures. Sure, you can optimize what the FZ1000 does by adjusting the exposure time, but sometimes I just don't have the time to do that.

jebarker 3y ago

Not only do phones have the upscaling blur, there's also a telltale "painterly" sign of convolutional neural network based processing that I see everywhere now especially in Google products.

MSFT_Edging 3y ago
It honestly makes some photos lose detail. Something you could make out in pixelated form just gets destroyed in the CNN processing.
- jebarker 3y ago
  That's an interesting observation that human vision can make out detail from a noisy signal. Non-generative CNNs are typically going to filter noise due to the loss functions used. Generative models could in theory learn to recognize whatever signal in that noise humans are queuing on and guess at the missing content. I think there'd be resistance to generative models in photography pipelines though based on the "made-up pixels" argument.

vel0city 3y ago

> automatically does the cool aperture settings for a foreground focus, blurry background, and other fancy things. Its not actually doing any cool aperture settings, its faking it all with software. And depending on the software stack and the picture in question, it can do a pretty bad job at it. I do appreciate the effects at times and use them myself, but on many shots I'll notice the depth of field effects being used in places where it shouldn't like too far on the shoulders, turning the edges of someone's hair to mess, a secondary subject next to but just slightly behind the main subject getting radically out of the fake depth of field, etc. Most of the photos I take day to day are on my phone's camera, for sure. But when I'm actually wanting to really take some pictures, I bust out an actual camera. My current main goto is my G9x Mark II, which is a lot of fun and can still send photos straight to my phone after shooting so I can easily send them to people right away.

MSFT_Edging 3y ago
Yeah I'm just talking about this at a high level. Its basically the same problem, so much auto-processing is done to make pictures generally look good, so that so many intentional pictures just look disappointing.

rob74 3y ago

Not wanting to carry an assortment of lenses with me, I have used bridge cameras for ages, and my current one (https://en.wikipedia.org/wiki/Panasonic_Lumix_DMC-FZ1000 https://en.wikipedia.org/wiki/Panasonic_Lumix_DMC-FZ1000) is pretty well-reviewed and generally does make really good photos, but my 6-year-old Pixel 2 phone still takes better HDR pictures. Sure, you can optimize what the FZ1000 does by adjusting the exposure time, but sometimes I just don't have the time to do that.