Monday, 3 December 2018

Week 4 [3-9.12.18] Improving image quality - DIY with Deep Learning


Hello, today I have a few articles about image processing for you:

https://venturebeat.com/2018/07/09/nvidia-uses-ai-to-clean-up-messy-photos/
https://arxiv.org/abs/1803.04189
https://www.engadget.com/2018/07/10/nvidia-ai-fix-bad-photos-deep-learning/

(I don't want to spark controversy. Why are so many articles here about controversial topics? I could have also gone for: https://www.engadget.com/2018/11/29/china-halts-gene-edited-baby-research/ - but I don't want to discuss ethics all the time).

1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

2. Do you use any methods to improve the quality of them? (yours or otherwise?)

3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?

4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)

38 comments:

  1. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?
    Yes I do like taking photos, very often the biggest problem is when my battery dies :P but of course the lighting problem is also very problematic one, You may have perfect frame, but it is super dark. I think that experience, knowledge and good equipment changes a lot.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)
    I certainly use photo processing software in post-production. Because who does not use them these days. I really want to make a very good photo that would not require any graphic work. It is all Adobe programs like Photoshop, Lightroom or After Effects.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    Of course, editing photos is just as difficult as taking a picture. It requires patience, many hours of learning and experience. It is not a solution for everyone, especially if it does not derive from pleasure. sometimes I use built-in applications, but let's face it, the quality of the pictures after using them has anything to do with a professional rotor, or even the use of ready-made presets.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    I think that you have to be a real passionate to go all the way to such advanced techniques. Personally, I do not think I would take advantage of it, but maybe I will be more interested in such a topic out of curiosity.

    ReplyDelete
  2. Thank you for sharing. Commercial image processing software is receiving the improvements found by science - I've seen a demo of some novel deblurring technique in PS once.

    ReplyDelete
  3. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    Yes! I really like to take pictures. Especially since good SLRs have appeared. Of course, often the images come out poor quality, basically always when you manually set the iris or time. That's why online preview is personally important to me. Sometimes when a solar image falls into the lens, the picture comes out marvelous in itself. However, there are situations when it obviously damages the picture taken.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)

    Usually I use Adobe Photoshop or GIMp's to improve image parameters, cropping or trimming. We don't always need an old car in the background of the palms or pyramids. I often make photomontages and then I use this software too. Red eye reduction or other discoloration, eg from the sun.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?

    I do not like using such "monkeyish" solutions. Everyone sees the colors differently, each photo wants to have something else. I try, if I change something, do it step by step. Using only and exclusively the single functions of more complex software.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)

    Cool, but too automatic for me. Though write such algorithms ... hmm a challenge! I will have to look closer and recognize the market in this matter. Thanks for the material.

    ReplyDelete
  4. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    I really don't...I heard before that we are generation that commemorates so many achieving so little. I totally agree with that!
    Like everyone from time to time I make a photo using smartphone and their quality is...not very good, but sufficient. Situation which are described in the article isn't uncommon.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)

    Only build-in programs in smartphone. I'm not a person which person which trying look perfect or get perfect photo every time. If photo is broken I just delete it.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?

    I'm worried about that. When I read article headers I think about it. That's really dangerous. I don't want to someone get any of my private inforamtion. I have all my photos on harddisk without internet connection (backup also don't have internet access). We can't predict how us pictures could be used...

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)

    I'm not really interested about trying any od that app...or maybe only to get know-how information how it works-maybe this deep learning algoritm could be use in other, more useful way.

    ReplyDelete
    Replies
    1. This is a nice quote about our generation :D

      About privacy - I completely agree - that's why I pointed to DIY AI, which can be applied without exposing the pictures.

      Delete
  5. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?
    Yes, I like taking photos but not a lot – only important moments. Unfortunately they are very often with some problems – in general they are blurred. Maybe it is so because of the quality of my camera.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)
    I’ve been never modifying my photos. I’m trying to focus during taking photos whether they are correct or good enough. If they aren’t I delete them and try to do it better. This is my method of improving my photos ;)

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    I really don’t trust such applications. First of all I do not have direct control over the privacy of my pictures. Secondly, I would like my photos still be amateur. Why? Because they are made by me only – not a professional photographer, and they show reality without any interference.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    Not really. As I told, it is not my cup of tea so I’m a bit sceptic in this topic. On the other hand I suppose that improving image quality can be very useful in criminal investigations.

    ReplyDelete
    Replies
    1. Fair enough. I also do very little (on average) to the pictures I've taken, but it's a good practice to always keep the originals.

      Delete
  6. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    I like to take photos and I used to take a lot of photos. Yes sometimes they are noisy, blured etc.Usually it is easy to take another better photo. Sometime this is not a problem but an effect that add to a picture some atmosphere or alluring effect.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?) In a times when photography was my hobby I used to postprocess photgraphies in a photoshop but usuall it was just minor corrections or effects. Usually it was easier just to take another photo than to improve the low quality one.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    I don't like the camera application with a button "Make me pretty", when i have used something like that the effect was "make me strange". I think that better is to take good photos instead of using apps.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    I don't think so. As I said before I enjoy taking good photos more than postprocessing.

    ReplyDelete
    Replies
    1. Again, that's quite commendable. But what will you do when time comes, there is no second chance to make a better picture - the person has died, or the sun has already set?

      Delete
  7. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    Yes, I definitely like my friend, a professional photographer, and I’m learning a little from him, and he helps me to photograph and process photos. This is a great hobby that helps to develop fantasy and a sense of ecstasy.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)
    Yes, I use many programs to process and retouch photos. Most often it is Adobe development tools. Also as a photo is very important camera. When I had a Canon 5D photo camera - the photo processing lasted much longer than when I bought myself a Canon 5D Mark II. In such cameras, they are already using software - photo enhancements, and using professional technology - you can reduce the photo processing time.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    Photo processing is a long and painstaking procedure. It takes a lot of patience.


    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    Unfortunately, I have little time to enjoy my hobby, but I’m happy to get acquainted with this material. Thank)

    ReplyDelete
    Replies
    1. If you're not using AI right now, it's probably coming to the Pro tools anyway.

      Delete
  8. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?
    I like to take pictures, but unfortunately I do it very rarely. I think the biggest problem with noise is when taking pictures in low light. Then it is often required to have a higher ISO setting, which can result in unsuccessful photos.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)
    Now you most often take pictures with your phone. I don't pay so much attention to whether the photo is perfect. If I need to make some minor corrections I use Gimp, but there is also an option to improve the photo directly from my phone or using Google services.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    As I mentioned, I use Gimp which is open source software. The fact that we do not know exactly how closed software works does not change what we get as a result. Often these algorithms are the dominance of one software over another, so we can not count on them being made public.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    I think it is worth knowing that something like this exists and it would be interesting to analyze how it works. However, I think that real enthusiasts of this particular field will take care of it.

    ReplyDelete
    Replies
    1. It's true - many algorithms are patented and kept secret. So any human is liable for using them. But what about AI? If it learns from examples, and does a similar job? Does it infringe on the patent then?

      Delete
  9. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    Of course , I do like taking photos, but the problem of our times with photos is that we mainly use mobile phones for that purpose. And of course phone cameras are getting better and better, but many pictures that look awesome on your phone screen are not that awesome when you look at them on your computer. There is still room for improvement.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)

    I sometimes improve photos using options available on Instagram or in Photoshop.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?

    I don't really trust applications which require access to my phone and personal data.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count).

    I'm not that into photography yet.:)

    ReplyDelete
  10. I don't take a lot of pictures, but when I do then may main problem is the composition rather than any technical issues which are handled by my camera.

    I can apply some cool, predefined filters in the camera application to improve my photos :) I also was "a master of Photoshop" in the past. I wonder why I didn't become a famous digital artist... I was also using some functions from OpenCV and my own functions to improve contrast of images for text recognition if it counts as improving photos' quality.

    Machine learning based solutions can be better than simple filters from a camera applications because they can be used to generate some data that is not included in the picture itself.So for example if there is a glare in the picture ML model can remove it and reconstruct adequate part of the image while filters could only use surrounding pixels to somehow interpolate values in the place where a glare, blur or something else is present.

    ReplyDelete
    Replies
    1. It definitely counts!

      Also, I admit that technical quality is trumped by image content and composition. And this is much harder to fix by AI (if at all possible).

      Delete
  11. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?
    Yes I do like taking photos and they are not perfect in some cases.


    Do you use any methods to improve the quality of them? (yours or otherwise?)
    I am using boomerang and pixaloop.


    How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    User experience is always one of the most important criterias in matter of choosing solution. So if the options are complicated sdk or library vs one click mobile application most user will always choose the app.


    After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    Yes, this topic seems to be really interesting and it would be great to play around with some specific tool.

    ReplyDelete
    Replies
    1. Interesting. I haven't found the boomerang thing, but pixaloop looks nice, although it looks like it's more CGI.

      As for the "users will choose the one button way" - maybe, but should they? It can be technically done keeping privacy, but I'm not sure if it always is.

      Delete
  12. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?
    Yes, I consider myself as a person who likes to take photos on some occasion. And as other, any other smartphone camera user I do sometimes experience problems that are rooted in hardware limitations like low lens quality, inadequate physical matrix size in terms of low light capabilities and grain exposition.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)
    I'm trying to take photos as best as its possible, but beside colour correction I am giving a chance to my photo being corrected by the build-in technologies and AI solutions provided by company responsible for building my phone.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    As I have said before I'm not a professional photographer and I'm not trying to find better software solutions for image improvement. Companies building today's smartphones and professional cameras spend millions of dollars in research, software development and hardware integrity to drag from matrix as much data as it is possible. By clicking "Make Me Pretty!" we are giving build-in AI systems an option to modify data provided by matrix in best possible way for this specific hardware and I can't imagine that some third-party app is able to do it better.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    No, I didn't use that kind of software – but I think It might be entertaining

    ReplyDelete
    Replies
    1. Well, it doesn't need to be as third-party as you think. The articles often come from people who work for the tech giants, and then do some publishing on the side. There are great things out there, just without the marketing packaging.

      Delete
  13. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    I am not so good in taking photos. I take photos when I go for a trip with my family and friends. I don't like take photos. Sometimes they are blurred.

    2. Do you use any methods to improve the quality of them?

    Yes, I do. I try to use the automatic method provided by a software to view photos.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?

    I use open source software and I think its a good idea to use them for this purpose. I think that we have to choose this method that better work for our picture oraz take less time.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)

    Yes, I do. It worth to try deep learning to correct pictures or manipulate them. It's very interesting topic.

    ReplyDelete
  14. I think we are choosing controversial topics so we can have strong opinions :-)

    1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?
    I like macro-photography. The main problem for small scales is depth of field, so only small portion of image stays sharp. Other aspects are easily controllable, as it's lab-like environment.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)
    I stack multiple exposures. After aligning and taking sharp portions of multiple shots, we can have stunning results.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    I'm all for open source, not using any closed tools for this. One can't say if it's better this way - the documentation might be lacking, and construction of processing pipeline requires much time investment. But I think it's worth, as I have full control and once set up, it needs no effort to use.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    This latent space thing is awesome! I'd like to try some deep learning structure-from-motion projects, as I'm fascinated by SfM for a long time now.

    ReplyDelete
    Replies
    1. Oh, yes. SfM sounds great - just record a movie of anything and - bang - there's the 3d model (or maybe not bang, but hours of computing time - I don't know). If you find something e.g. on GitHub, definitely throw me a link.

      Delete
  15. 1. Yes, I like taking pictures. Currently, many people take pictures using a mobile phone (including myself), for my needs it is quite sufficient, you can often get very nice results. The big downside I noticed when taking photos with a phone is that they sometimes come out fuzzy, because when I take a picture I shake my hand and it’s hard to catch the focus then.

    2. Mostly all kinds of filters offered by Instagram and Snapchat.

    3. If I would like to make a big change or improve my picture, I would prefer to do it in a program for this purpose, for example, Photoshop. Single buttons that improve the entire picture all the time, always fail and the end result looks not like I expected and sometimes comically.

    4. I have never tried something like that, but after watching the movie it seems like a great fun. Maybe in the future, I will try out TL_GAN. Thank you for showing an interesting image processing option :)

    ReplyDelete
    Replies
    1. You're right. The phone as camera has failed me mostly, when I wanted to take a picture quickly.

      Delete
  16. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    I really like taking photos and I like to be photographed! My camera is not very good, so sometimes the photos I take are blurred, especially when it is dark. It is difficult to stabilize the camera and to obtain a photo of good quality. Another problem is related to sharp shadows: it happens when the day is sunny, so wide dynamic range can cause half of my face to be completely black.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)

    I usually take photos on my mobile phone camera to send photos to my friends via messengers. In this cases I do not improve the quality of them, just take and send them immidiately. I use only my mobile phone camera even when I go for holidays. I take lots of photos to have larger to set choose from. Then I pick the most impressive of them and try to improve their quality using Photoshop software.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?

    I think that for the people, who do not have time or are not very good at improving the quality of photos, the option like “Make Me Pretty!” seems to be quite useful. If this functionality is properly programmed works in a stable way, it will become very popular and used by many people. You should have a great experience to achieve a good effect improving photos. Otherwise, it will be like with some stars – for example when they try to ineptly improve the photo.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)

    In my opinion it is very interesting. Deep learning is very popular topic and it can be applied in solving many problems also in image processing. The application presented in the link is quite interesting and fairly promising, it is very exciting to see how we will look like in x year. I treat these kind of applications as entertainment. I am more interested in the deep learning methods in image processing than use interface of application where these methods were applied.

    ReplyDelete
    Replies
    1. Well, sometimes an application of DL is not "important". But having fun creatively can sometimes have unintended positive consequences.

      Delete
  17. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    I like taking photos, especially on holidays (new fantastic places or events), but I'm not a photo geek. Of course my photos frequently have some problems like one you mentioned, but my strategy is to take a lot of them and drop the one that are not good enought. Usually works.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)

    As I mention, I am doing a ton of photos and after that I am filtering the best ones, but I'm not using any advance techniques, only those, that are implemented on my phone or on my camera.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?

    As I mentioned, I'm not a geek, so usally I want to have just a photo, without any art, so I trust my smartphone. Only sometimes before I buy new one I reed some recomendations about which one is taking best photos, but it is only one of many criteria.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)

    I only used GAN (Generative Adversarial Network) in my work, where we were tring to help one of e-commerce companies to autmatically create some advertising spots and this type of deep learning perfectly fit this use case. Generally idea behind this type of neural networks, where we have one network trained to generate pictures and second one to discriminate not real images from real onces, is brilliant.

    ReplyDelete
    Replies
    1. That reminds me I have never played with GANs on my computer. My other advisor has actually suggested trying to add them to my research.

      Delete
  18. Thank you for mention that article, nVidia achieved a fantastic score in image processing. Noisy/blurred/incomplete photos are a common issue in medical imaging, and such algorithms could lead to improvement of speed of diagnosis (because there would be no need to repeat given examination). I'm often using social media apps, such as Instagram, to enhance my photos- mainly for light and contrast correction. About the applications, I think that we have to distinguish our audience. Of course "general user" would be interested in single-click solution rather than deep understanding of them. For programmers, digital image processing would be related to open source algorithms and libraries that you'd mentioned. And, for artists, it would be a handcrafted image parameters enhancement provided by graphics editing software, such as Photoshop or Lightroom. From my experience, I'd used "dlib" C++ toolkit with OpenCV to process digital images (and it works just great!). PS does anybody remember Prisma app?

    ReplyDelete
    Replies
    1. No, I don't remember it. OK, if I googled correctly it's a iOS style-transfer app. Thanks for mentioning Dlib - I'd have another look at it, because I only remember we used it as a black box for face detection. I've heard the code is especially interesting inside "D" in Dlib stands for "Design by Contract" or some other programming technique.

      Delete
  19. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?

    If it comes to taking photos, I am an amateur. I like taking photos, but sometimes I have a problem with lighting, focusing, objects in motion.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)

    I use popular graphics programs to edit photos, videos and music. An example: CorelDRAW, GIMP. Before taking a photo, I try to choose the optimum camera settings.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?

    In my opinion, this approach is useful. However, the final effect is important. For users like me, this is quite enough. However professionals, would prefer to use their own proven methods.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)

    I haven't tried these things yet, but I am interested in. I've heard about similar functions in program Adobe Photoshop, but I'm not sure.

    ReplyDelete
    Replies
    1. Not so specific, but the quality filters are already quite advanced. Maybe some day there will be "auto add smile" just like "auto red eye removal".

      Delete
  20. 1. Do you like taking photos? Are they sometimes e.g. noisy, or have any other problems (blur, under/over-exposure etc.)?
    Yes, I like taking photos ;) in fact there was a time, when I thought about taking photos professionaly and I still find a lot of joy in taking pictures. I don’t have any problems with the blur or the over/under-exposure, but of course I met some type of problems. I think that the main problem is to use a good light – whereas during using the professional equipment this problem is minimalized, in the proccess of taking photos by mobile phones it is way more difficult.

    2. Do you use any methods to improve the quality of them? (yours or otherwise?)
    While I am using my own equipment, I am especially focusing on the right photographic len- it can really make a huge difference. What’s more, I take time to adequatly frame the picture – I find the main points and decide where such points should be situated on the photo. I also care for the perfect lightining. I am careful not to blur the picture and catch the right amount of sharpness.

    3. How about using open sourced or published methods for such processing? Is it better or worse than just trusting your smartphone camera application, which has one button "Make Me Pretty!", that we don't know the workings of?
    When speaking of open sourced or published methods for processing the photo, I like to use Photoshop and some free equivalents. It is really great program which can visibly increase the quality of the photo. You can change the contrast, the sharpness or cut some space of the photo.. It is for sure better than to just trust smartphone camera apllications and using programmed filters. They all are basic and not personalised for each photo, therefore you can not gain the same effects as after using Photoshop or other programs to make over the pictures.

    4. After taking a look at this, do you have an interest in trying out some of the deep learning image processing available e.g. on GitHub? Or maybe you have tried it already and can share a story? (Things like TL-GAN: https://www.youtube.com/watch?v=O1by05eX424 also count)
    A few years ago I used openCV and face recognition.
    Now I really like real time object detection, which is presented on the webpage:
    https://pjreddie.com/darknet/yolo/

    ReplyDelete
    Replies
    1. Same here, I also play with YOLO, I just didn't fully upgrade to YOLO3 yet.

      I agree that some things work best after the photo is already taken. When taking the photo it's good to focus on the composition, perspective etc. If other things can be fixed later on, why not fix them afterwards?

      Delete
  21. I like taking photos, it has become extremely easy to make them in quite good quality since we have smartphones. I try not to get used to the idea they have to be perfect, because I am afraid of the situation, when I’ll concentrate on the technical aspects instead of visual impressions. So, I have no problems, even when they are not ideal.
    That’s the reason, why I don’t use special methods to improve my photos. The only one is to make several pictures, compare and choose the best one :-)
    Now I can see, that I have answered this question already, because for my individual use, smartphone’s application is quite enough. But on the other hand, I think, that working on technical side of the pictures is some kind of art, because this way we can provide better impressions. One photo can be presented in many different ways, colors, using sharpness, brightness/darkness etc. It can give some fun during such creative work and I can understand people modifying pictures for some kind of purposes, professional ones in particular. But to use it only to amend reality, seems to be not enough reason for me. I’d rather remember real world instead of artificial one.
    Deep learning image processing could be interesting, because the result may be surprising, and often it is. I have no exact example, as I am not passionate of such techniques, I’m afraid. But you gave me some inspiration to check this issue, thank you.
    BR, Marta

    ReplyDelete