Upscale your Images using DEEP SUPER RESOLUTION with ESRGAN

Hi Nicholas, Do you think it is scientifically meaningful to enlarge different types of images using SRGAN trained with a different dataset?

since you did this video a year ago, would you say a year later that ESRGAN is still #1? If not, what would you recommend?

Can I modify the same code for satellite imagery?

Beautiful

Niye bizden kmse araştırmamış?Çünkü kapasite yok.

Thank you so much! I needed it in my project, and now I know how to do it!

Thanks man!

Is there a cleanup-only option? The X4 .pth file leaves a checked pattern all over the picture and does not clarify the details of real pictures. Great video.

okay, let search for a "low resolution image" and then google, "small images"... :D :D :D

Great stuff. Thank you so much

Why my cmd do not recognise the git keyword while cloning

Unfortunately, ESRGAN can only handle small size images. It will not process any image larger than 1k. So clear limitation there. Good tutorial there Nick!

Hey Nick, can you please tell me how to run this on google colab?

Great Sir

bro thanks thanks to you my task is done, big thank you
you are amazing

Can you also teach us to train custom dataset on ESRGAN

Hello i am using a macboook pro and then torch 2.0 is the avalaible version so i can't hahe cuda

You should take 4k image resize it to 640x480, do upscale with this AI model and show us the result. We want see the difference. It works not as expected.

I wish you implement it instead of sharing the code only

Thanks for this tuto helped me a lot, even if i cant deal python im very very beginer and my english is medium so ... anyway its working for me with on windows but ... i still have a question , when i download the others models they are not working, the file is in "pth" format does that mean "pth" models work with python only ?

Also do you know how to use different size of upscale without python like x1 x2 x3 i tried but it looks like its not working " it working only on animevideomodel x2 & x3 yes but unfortunately no x1 " do you have any information about that thanks a lot in advance

Hi there, I have an error, it writes: CUDA out of memory. Tried to allocate 4.69 GiB (GPU 0; 8.00 GiB total capacity; 6.23 GiB already allocated; 0 bytes free; 6.53 GiB reserved in total by PyTorch) If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF. Do you have any idea? Do I need to install virtual environment? Thanks

Much appreciated

well i used to think size doesnt matter

Can you make a video of image super resolution using swinir transformer?

Is there a way to tune this model?

Looks really easy to do, now I wonder if it's possible to upsample a bunk of images, for example the FER2013 (which is 48x48)

I am facing the following problem. My laptop does not have a GPU, when installing pytorch in Compute Platform I choose CPU. But when I run test py I get the following errorI am facing the following problem. My laptop does not have a GPU, when installing pytorch in Compute Platform I choose CPU. But when I run test py I get the following error: Torch not compiled with CUDA enabled. How to fix it?

Please full Tutorial on GFPGAN-1.3.8 with scratch

Does it work with videos?

Thanks so much for the tutorial, dude. You gained one more subscriber.

ERROR: Could not find a version that satisfies the requirement torch (from versions: none)
ERROR: No matching distribution found for torch

You seem like a smart person but I am surprised at how you compared upscaled images. It's not just about resolution but fidelity, you can increase the resolution of any image using even Paint, it's the ability to reproduce missing details that makes AI super resolution interesting and this model fails to do that.

If reserved memory is >> allocated memory try setting max_split_size_mb to avoid fragmentation. See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

Brother please help.

should try it out on images of UFO or Loch Ness Monster :)

I've come to the conclusion (being a bit cynical) that the best results are obtained by taking a good photo, then reducing it in size, and adding a blur, then put it back through these (various) engines to obtain what you started with. On the other hand, with a genuinely originally bad image, with all the personal photos I've tried, of various quality and size the results are worse (much worse) than using the good old paint shop 5 in built resizer (that presumably interpolates, and does so very well). Or failing that just using a sharpen option.

When you gonna to do a tutorial json files openpose to bvh?

thumbnails

I'm a newbee and I was wondering how I can approach a problem where I have to assess the damages (tears, scratches, spots etc.) of old photos.

Congrats on hitting 100k subscribers earlier this morning!

not having any gpu. i installed pytorch on cpu, and it throws an error. any help.

What do you need to install before cloning the github?
When I try the git clone I get this error:
'git' is not recognized as an internal or external command, operable program or batch file.

I typed "git clone" but received a message from the command prompt (running as an admin) that "git" is not a valid command.

Edit: I ended up just downloading the .zip and extracted it anyway, just curious why I couldn't use "git" on my PC.
Edit 2: I get the same issue with "pip3", is there some sort of prerequisite for these commands to work?
Edit 3: OK, I'm making these edits incase someone else has the same issues. I was able to get "pip" to work after installing Python directly from their website AND did a restart on the PC. It's downloading Pytorch now.

Last Edit: Awesome tutorial! Your methodical way of explaining things has helped me keep up with what you're saying and doing! I was able to get this working and it is really cool to see just how far we've come in terms of computing abilities.

Great project! But the upscaling amplifies JPEG compression artefacts, so better use lossless formats instead, such as PNG. Maybe we need a model that was trained with those artefacts and "knows" how to deal with them!

Is it just me or it not working very well with group photo (like 50 people)... ... All human faces distorted

great tut

My issue is most of these pre-trained models smooth out images losing their inherent, interesting and important details.

PyTorch CUDA works in amd GPU?