Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

How My Images Are Dithered

dead.garden

Post reply on HN
Print view View original

71–80 of 81 posts

Re: How My Images Are Dithered

#71

Post by ipunchghosts »

dahart wrote: post #52
ipunchghosts wrote: post #12

Why is this not a typical data augmentation method for training deep networks? It would allow the networks to learn invariants that's align with humans. Currently, I suppose none of these dithered images would be correctly classified.

Perhaps because inference on dithered images is pretty rare, not a common or popular thing to do? Otherwise, people would train on dithered images. NNs are trained on noisy images. Training with dither would not learn any other invariants besides how to see through dither, it doesn’t automatically generalize to anything else.

I am not sure that's true. It would learn to not focus in texture which is what current nns do.

Re: How My Images Are Dithered

#72

Post by ipunchghosts »

trollbridge wrote: post #19
ipunchghosts wrote: post #12

Why is this not a typical data augmentation method for training deep networks? It would allow the networks to learn invariants that's align with humans. Currently, I suppose none of these dithered images would be correctly classified.

I’m sorry, but I can’t understand what you’re saying at all. Explain?

There are many features in an image dataset which a can correctly classify the dataset. Neural nets usually focus in texture which we human clearly don't, we focus on a higher level abstraction and usually ignore texture. Using either would encourage the network to use broader features and ignore texture

Re: How My Images Are Dithered

#73

Post by ipunchghosts »

dist-epoch wrote: post #33
ipunchghosts wrote: post #12

Why is this not a typical data augmentation method for training deep networks? It would allow the networks to learn invariants that's align with humans. Currently, I suppose none of these dithered images would be correctly classified.

When someone says AI can't do something, I like to try it: > This image features a close-up abstract scene styled with a classic halftone (Ben-Day dot) pop-art filter ... https://share.gemini.google/O5eqZOKxwzw3 > This image is a halftone/thermal print styled photograph. The photo features a group of four people posing closely together indoors or in front of a windowed backdrop: https://share.gemini.google/KJOxBDXYsF…

This is a good way to think.

Re: How My Images Are Dithered

#74

Post by zahlman »

rahimnathwani wrote: post #9

If you're doing halftoning with the intent to get custom DTF (direct to film) prints for t shirts, you might find this tool helpful: https://tools.encona.com/dtf It aims to 'knock out' some areas of the image: make them transparent in the digital image so those areas are not part of the final film. This means less of the surface of the t shirt has film on it, allowing it to be breathable and flexible. I got some DTF…

> DTF (direct to film)

Very important clarification, there.

Re: How My Images Are Dithered

#75

Post by zahlman »

galleywest200 wrote: post #24
ReactiveJelly wrote: post #22

> Dithering can reduce file size Yeah, if you're using lossless compression like PNG or GIF, adding a lossy step to reduce the number of colors in the palette can help. If you're compressing photos, you should use JPEG or WebP. Anecdotally, JPEG usually beats dithering, and WebP always does. It's fine to dither images but please please admit that you're doing it for the aesthetic. WebP has very broad browser support…

> It's fine to dither images but please please admit that you're doing it for the aesthetic They did in the article. > My goal was to immitate a printed image.

Sure, but offering a weak file-size justification alongside that confuses the message.

Re: How My Images Are Dithered

#76

Post by zahlman »

mort96 wrote: post #41
ReactiveJelly wrote: post #22

> Dithering can reduce file size Yeah, if you're using lossless compression like PNG or GIF, adding a lossy step to reduce the number of colors in the palette can help. If you're compressing photos, you should use JPEG or WebP. Anecdotally, JPEG usually beats dithering, and WebP always does. It's fine to dither images but please please admit that you're doing it for the aesthetic. WebP has very broad browser support…

.. What do you mean by "beats" here? Is the article trying to hide the fact that they're doing dithering for, among other things, artistic reasons? A dithered image almost certainly beats a non-dithered JPEG compressed image in terms of file size, because you can JPEG compress the dithered image. Thus, if you like the look of a dithered image, you get file size reduction as a bonus. I swear, some people on HN intenti…

> A dithered image almost certainly beats a non-dithered JPEG compressed image in terms of file size, because you can JPEG compress the dithered image.

For an algorithm like JPEG, if anything, you'd need to blur rather than dithering.

Re: How My Images Are Dithered

#77

Post by dahart »

ipunchghosts wrote: post #71
dahart wrote: post #52

Earlier quoted context omitted.

Perhaps because inference on dithered images is pretty rare, not a common or popular thing to do? Otherwise, people would train on dithered images. NNs are trained on noisy images. Training with dither would not learn any other invariants besides how to see through dither, it doesn’t automatically generalize to anything else.

I am not sure that's true. It would learn to not focus in texture which is what current nns do.

Which NNs and what tasks are you thinking of? DLSS, for example, reconstructs texture, along with other denoisers, as well as all video models. Are you thinking only of object recognition?

Why do you believe training on dithered images would learn how to see through any other kinds of image modification? In my experience, image NNs learn only what you explicitly show them, and they don’t learn concepts they’re not trained on. All the denoisers I’ve ever seen have trouble handling subtle changes in pixel filtering they haven’t seen in the training data, and they don’t easily learn how to denoise things like clouds or reflections. I’d be willing to bet that if you trained on halftoned images, NNs would not be able to handle Floyd Steinberg dither. They just don’t generalize automatically.

Re: How My Images Are Dithered

#79

Post by nicbou »

inigyou wrote: post #61
trentor wrote: post #27

I was a rotogravure printer in another life, and this post made me a bit nostalgic. In intaglio printing you usually don't see these patterns because the electrostatic assist pulls the ink out of the cells, and the colors blend together more. It's more akin to a developed photo. I mainly remember them from having to do offset printing for the intermediate exam during my apprenticeship. Our paper was about 3.5 meters…

I can imagine that feeling of "being a cog in a beautiful machine" gets a lot less people these days because so many cogs are automated and so many machines are no longer beautiful.

On the other hand I vaguely remember a comment from a printer who was radicalised by the amount of paper wasted on printing flyers, day after day, week after week.

Re: How My Images Are Dithered

#80

Post by ipunchghosts »

dahart wrote: post #77
ipunchghosts wrote: post #71

Earlier quoted context omitted.

I am not sure that's true. It would learn to not focus in texture which is what current nns do.

Which NNs and what tasks are you thinking of? DLSS, for example, reconstructs texture, along with other denoisers, as well as all video models. Are you thinking only of object recognition? Why do you believe training on dithered images would learn how to see through any other kinds of image modification? In my experience, image NNs learn only what you explicitly show them, and they don’t learn concepts they’re not tr…

Eero simmoncellis group showed that specific network modifications allow the networks to extrapolate beyond noise they were trained in.

NNs can certainly learn beyond what they were trained for as new architectures have demonstrated. See benchmarks for any if the dan hendryks imagenet corruption dataset benchmarks (imagenet c r and a). Data + architecture work together to enable.

I have work using neural networks trained on synthetic aperture sonar SAS imagery using language only. The SAS datsets are extremely hard to come by. The dataset I used was from parts if world I can assure you we're never in the training set. My approach exceeds what clip models can do zero shot wise. Therefore, the network classified imagery if a modality and environment 100 percent not guaranteed to be in the training data.

Post reply on HN