jjl

Lives in United States Portland, OR, United States
Works as a office slave
Has a website at www.phlumf.com
Joined on Apr 8, 2003
About me:

10D & 20D
whole slew of lenses...

Comments

Total: 32, showing: 1 – 20
« First‹ Previous12Next ›Last »
In reply to:

entoman: That's quite clever - but basically just filling a defined area with a photo-realistic texture. Nothing very new, but faster and perhaps more convincing than doing it manually in photoshop.

That's not really what it's doing at all. I suggest doing a deep dive in to what Machine Learning is all about. It's going to fundamentally change a lot of things. This is just the very start of it.

Link | Posted on Mar 25, 2019 at 05:47 UTC
In reply to:

Leicalika: what's next; a pick-up line generator for photographers?

"say miss, you shoot raw"

Hey, baby... do you believe in f8?

Link | Posted on Mar 25, 2019 at 05:45 UTC
In reply to:

Guenter Hofstaedter: Nvidea turns artists into workless people

It's going to change art, but not likely replace it - just bring it to another level.

You can bet that an artist using this technology will be able to create far more interesting things than a regular person. Now, they'll be able to do it that much more quickly.

Link | Posted on Mar 25, 2019 at 05:44 UTC
In reply to:

Jakub Kubica: Those images looks way off. Its like drawing os mentally ill person, look at the tree from example... sure it contain textures etc that are "realistic" but whole composition gives me headache for some reason. When you look at the tree, it does not make any sense to have that shape - also - no shadow.

What's amazing is that it was able to come up with a nearly nonsensical tree to fit the shape that was drawn. The people drawing those shapes are just random people. Give this technology a few years, and put it the hands of an artist. I think you'll be shocked.

Link | Posted on Mar 25, 2019 at 05:42 UTC
In reply to:

Mark Turney: Another “solution” looking for a problem. What a waste of time.

On the contrary, this is going to fundamentally change many technologies and fields of work. This is just the most visible aspect of it. Machine Learning algorithms are an entirely new way of solving problems, and the real-life applications are just getting started. This is like the early days of the computer, all over again.

Link | Posted on Mar 25, 2019 at 05:40 UTC
In reply to:

Fixx: Can it draw naked women?

short answer is - absolutely. Someone just needs to create training data for it. What you need is about 100,000 images of naked women, and someone would have to analyze them all to pick out various characteristics of them.

Link | Posted on Mar 25, 2019 at 05:39 UTC
In reply to:

Roland Karlsson: The results look weird in one way or another. It do not really understand how a landscape should look. It makes a rather good try though.

What you're seeing now is like the drawings of a 5 year old. Not great, but it'll get better. In 10 years, this technology will be producing far better results. Plus, the detail and scale on these is essentially infinite. If you can render a tree, you can zoom into the leaves and render the macro scale too.

Link | Posted on Mar 25, 2019 at 05:37 UTC
In reply to:

mikeoregon: Considering that this is a first try, the output is impressive. Think what these images may look like in 20 or 30 years. Also, consider what will happen when AI-generated figures are animated inside the AI-generated landscapes. We're likely to see films that are completely computer created and we won't be able to tell.

This is exactly where this is headed. At the moment, there are strange fractal-like artifacts in the transition zones and details many of these images, but that'll just get better with time too.

In addition, synthesizing voice and sound with machine learning is just as easy (if not easier) than with images. Even writing styles can be learned this way. You could describe your dreams, and see a movie of it (machine-generated script, video and sound)... granted, not likely a good one, but with a little human tweaking? it'll get better.

Link | Posted on Mar 25, 2019 at 05:35 UTC
In reply to:

BodkinsBest: So it analyzes a crude drawing, goes online and steals photos to butcher and recreates the crude drawing with the bloody pieces? How lovely...

The images for the training are not stolen. They're generally licensed under Creative Commons. This kind of technology doesn't need good quality photos, just bulk. All kinds of non-photographers share their images on Creative Commons licenses - think your average Joe.

Link | Posted on Mar 25, 2019 at 05:32 UTC
In reply to:

PostModernBloke: It's simple, the images you create are your property, or the property of the client which commissioned them; depending on the contract.

Don't upload them - they'll be nicked.

The images being used for these purposes aren't coming from professional photographers. They're coming from your average person with a cell phone, who doesn't really care if their photos being used under Creative Commons licenses, and will never see their images in some unexpected place. There's no need for professional-quality photos for this kind of use. it's all about quantity and variety - not quality.

Link | Posted on Mar 16, 2019 at 17:55 UTC
In reply to:

jjl: I'll go against popular sentiment on this one.

Copyright is there to prevent copying of your images. People using these images for machine learning don't need to copy your images. If you've uploaded your images to their server, you've already done the copying. All they are doing is looking at the images, creating their own tag data, and then using a combination of the image and the tag data to generate an algorithm. What was unique about that image is completely lost in the process. There's no way to re-extract the image's data from the result.

To me, the people who do this are making a derivative work, and that should be allowed.

If you remove your images from the server, they'll no longer be available - which is how it should be. If you look at these creative commons data sets, there are missing images, due to this reason. They have to constantly refresh it.

Also - not that it matters for legal reasons - they don't need quality images for this, just quantity.

@3pgrey But, what if they're just using the images to "look at them" and spur their creative process? To me, that's essentially what the ML users are doing. They're not reselling the images. You could make the case that they're indirectly profiting from work that required your images as one input... but that's the case with any work - it all relies on a myriad of input.

also, just to note... nobody is going to hire a photographer to shoot images for purposes of ML. It's completely impractical. You need thousands and thousands of images of similar objects in all different kinds of situations - and most of them are crap for any other use (think - random people's cell phone pix). Might it be possible for someone to have a business like "sell me all your duds for a penny each"? possibly. But, there's no reason for anyone to hire skilled photographers for this kind of work.

Link | Posted on Mar 16, 2019 at 17:48 UTC

I'll go against popular sentiment on this one.

Copyright is there to prevent copying of your images. People using these images for machine learning don't need to copy your images. If you've uploaded your images to their server, you've already done the copying. All they are doing is looking at the images, creating their own tag data, and then using a combination of the image and the tag data to generate an algorithm. What was unique about that image is completely lost in the process. There's no way to re-extract the image's data from the result.

To me, the people who do this are making a derivative work, and that should be allowed.

If you remove your images from the server, they'll no longer be available - which is how it should be. If you look at these creative commons data sets, there are missing images, due to this reason. They have to constantly refresh it.

Also - not that it matters for legal reasons - they don't need quality images for this, just quantity.

Link | Posted on Mar 16, 2019 at 06:01 UTC as 11th comment | 7 replies
In reply to:

Alphaloki: "After finding your anchor points and running the cables between the two points, the system is tightened using a winch – no tools required." Isn't a winch a tool?? I mean, I've never seen anyone pull a winch out of their ...ummm, shirt. :)

The winch is likely built-into the cable... so it's part of the system, not a separate tool.

Link | Posted on Sep 15, 2016 at 04:49 UTC
On article Striding Forth: Canon EOS 5D Mark IV Review (2154 comments in total)
In reply to:

DouglasGottlieb: Refocusing from tip of the eyelash to the pupil on 85mm 1.2 shallow DoF head shots would be super useful.

I wonder if this would be more significant using super-macro lenses, where the DoF is miniscule at just about any aperture. Could this be used to effectively double the depth of field on those shots? I do a lot of focus stacking, and this would be interesting to try. I don't think it'd save any memory, as these 2x-the-focus shots would be 2x-the-size. Anyway, it'd be fun to try... especially with the MP-E 65mm.

Link | Posted on Aug 26, 2016 at 05:38 UTC
On article 6 things iPhone photographers want from Apple (66 comments in total)

No. More. Megapixels.

It's a total gimmick. Most people only share their iPhone images electronically anyway, and shrink their images down to under 1MP. In addition, these sensors are so dang small, the individual photosites are laughably small.

Just give me better image quality (which you can do, by not having such ridiculously small photosites)

8MP, heck even 4MP is more than plenty for anything you need to do with an iPhone.

Link | Posted on Sep 9, 2013 at 20:56 UTC as 25th comment

Personally... I have a facebook account, but rarely log-in. I only have it because friends (actual friends) of mine will post things there, which I want to see, and I need an account to do it. (maybe once every 1-2 months)

Just remember - log out as soon as you're done. Some people think just closing their browser will log them out. Not so. Same deal with LinkedIn and many other sites.

Anyway, I don't think it's a big deal having a facebook account. It just depends on how you use it. They can't abuse your personal data if you don't post anything.

Link | Posted on Sep 9, 2013 at 16:03 UTC as 10th comment
In reply to:

Vitruvius: You still need 2 devices, so how does this solve anything? You still need to carry your smart phone AND this 'camera' without a screen or controls.

You save a bit of money (on rapidly outdated technology) and in exchange you are very limited to who, how, and when you can use it.

What is the point? What was the problem they are trying to solve?

Because most people already have their smart phone with them... regardless if they have another camera or not. This essentially gives you the option of a much better lens on your smart phone.

The smart phone also gives you much better/easier options for sharing your images... this setup makes it easier to share better photos.

Link | Posted on Sep 4, 2013 at 17:45 UTC

I've been asking for just this kind of thing for a while now... If they market it well, and the app/software is good, this could be wonderful.

Though, you could get similar results with a more traditional point-n-shoot camera that had good wifi connectivity to phones.

Personally, I love my DSLR, but actually use my iPhone more often, simply because I have it with me all the time. My DSLR is bulky, but the camera on my iPhone is limited.

Link | Posted on Sep 4, 2013 at 17:43 UTC as 59th comment | 1 reply
In reply to:

JDThomas: Are people actually READING the article? It's about Anthony Karen's methodology on photojournalism and trust. The subject is beside the point.

@Ryan... I hate to be the grammar police, but if you want to make a cogent argument, you need learn how to use English properly.

Personally, I think sunshine is the best antidote to hate groups. Give them a platform, and they'll show the world how idiotic they are. The photography excellent as well.

Link | Posted on Aug 15, 2013 at 15:37 UTC
Total: 32, showing: 1 – 20
« First‹ Previous12Next ›Last »