Deoldify on Windows...Help Please!

[Joey Bagodonuts]

Hi all,

I am looking for a brave soul who would be kind enough to offer detailed instructions on how to run Deoldify on Windows. I have downloaded Anaconda and Python, but can't seem to fill in the missing pieces to get it all to work.

Although you can upload small samples to the Google Colab it is hosted on, it would be great to experiment without the restrictions of file size and uploading and downloading online.

Deoldify can be found here: https://github.com/jantic/DeOldify
And the Google Colab for video colorizing is here: https://colab.research.google.com/github/jantic/DeOldify/blob/master/VideoColorizerColab.ipynb

If there is anyone who would be willing to take a few minutes to write down and post an installation guide on how to run this in a Jupyter notebook on Windows, I think you would be helping more than just me...but also a bunch of fellow enthusiasts who love working with this stuff, but who are not programmers!

Cheers,
Joey

[october262]

says here - https://alternativeto.net/software/deoldify/?platform=windows it's not available for windows.


also see here - https://colab.research.google.com/github/jantic/DeOldify/blob/master/VideoColorizerColab.ipynb

[Joey Bagodonuts]

Thanks for that link...

As I alluded to in my OP....you would need to run it in Anaconda, which I believe would be as a "shell" in Windows.
I am looking for how to go about doing that.

I was hoping someone who is familiar with the scripting in that environment could offer assistance.

[the_man_one]

cgywin and qemu. but it will not be native, those programs emulate Linux so its running on a distro inside an emulated environment. if you can spin up a vm in your win install like hyper v (win pro) you can do the same thing.

[Joey Bagodonuts]

Thanks for your reply, the_man_one....but the reason I am asking for assistance is because I do not know how to do it.

I am on another forum where I am, for lack of being modest, a solid expert. I get newbie questions all the time, and I can either answer them directly or not. But if I answer the question, then I answer the question first and foremost and not offer anecdotes, opinions, political views, and weather reports. So many people on these forums redirect from the OP into discussions about what they would do themselves. I get it...it's hard to have an organic "conversation" in forums.

So I am at a major disadvantage. I understand that here. I have tried for a week or 2 to try and grasp this whole "virtual shell" thing. It's not sinking in. So....I figured I'd throw it out there for anyone that might be willing to help on this.

If someone willing to help with an instructional on how to get Deoldify to work in Windows via Anaconda (or whatever) - then that's great. And I think that person will wind up helping a lot of people who want to experiment with this program without having to rely on the Colab. If not....I understand that too, which is unfortunate...but it is what it is.

Thanks to all of you who help the newbies. We all start out as them at some point. I try to always remember that. Enjoy & Peace

[aedipuss]

if you're willing to think sideways for a minute, why not install linux and dual boot? plenty of tutorials online.

[the_man_one]

ok if you have win 10 pro look up hyper v and do a read on it. too much to try to explain. basically you pull it up and install a Linux distro on it (this is an over simplified explanation. vm ware is another possibility. alternatively you could run a Linux hypervisor such as robo Linux or proxmox or kvm or xen or unraid or many others but if you did that just run a Linux distro and install it on there. there is no "easy" way to get it on windows. you could run something like codijy or black magic or akvis colorage natively in windows. those do photo things altho im not sure if they are what you need

[the_man_one]

what is going to happen is first establish which windows you have. if its win 10 pro then you have hyper v already there (or you can get it). the next thing is to determine if your processor will handle running it. then set up your "environment" which is called a virtual machine it literally is a fake computer you built inside of your os with virtual hardware. from this point to build it you have to allocate ram and core/s to your machine to get it to run and a hdd or portion of a hdd. then install the os on that drive.

I hope this is what you want. now if your windows does not have hyper v on it you have to go to an external app to do this (at this point you have to decide which you want) then do the same things I just said.

I tried to make it as layman as I could. reading on VM's and which "VM maker" (shell) you choose will help a lot. ill say this, the better processor you have the better this will do, the more ram you have same thing, and the bigger hdd or hdd's the better. im thinking minimum ram is 4 gig but 8 is recommended and I would not try it unless I had at least 16. the reason being you will have to run 2 pc's on your pc. a dual boot or even enterprise class hardware (along with the OS to back it up) is your friend in this situation. for instance I have multiple servers running 96 ram on 1 with 16 cores. I have another server with 64 cores and almost 1 tb ram. they run multiple flavors of Linux at the same time and multiple windows environments also. sorry im not concise with my explanation but there is a lot to learn and you have not told me much to go on

[the_man_one]

as a start you can look up home lab setups, networking, and hypervisors when you can understand those then maybe you will get a better understanding of what you want. believe me I know exactly where you are. if you want to do this do not give up its well worth the "play time" you get, at least it is for me.

[the_man_one]

Anaconda is a free and open-source distribution of the Python and R programming languages for scientific computing. it is not what you are referring to as a "shell" its a programming language for making apps I assume. I know nothing about coding or scripting. that's not what you want to do unless you are planning on taking raw code and make your own custom program to communicate between windows and Linux.

[Joey Bagodonuts]

Thank you for all your pointers, "the_one_man!" I will take a good deal of Saturday looking into it!

I am running an i9-9000 Intel with 64 GB RAM and an NVIDIA GTX 2060. Hopefully enough horsepower.

Thank you again...I will report back!

[the_man_one]

yes you have the horsepower for it. its a different kind of horsepower but it will work. enterprise class things are for working, sustained usage and such. your horsepower is akin to getting a '70 car dropping a big engine in it and not quite building it fully, ie: you left out the porting on the heads, over-bored the cylinder walls, and left the stock headers on, in other words it will run like a scalded dog but its going to run hot.

[the_man_one]

forgot 1, docker, look it up. it is quite possibly the easiest way using windows only, you do however need a minimum Linux kernel.

[Joey Bagodonuts]

I will look into it....

The creator of DeOldify suggests I run a dual boot of Windows and Ubuntu...he says my greates chance of success is to simply run it in Ubuntu...problem is, my system is UEFI and I am not sure I can arrange for a dual OS boot with that.

[the_man_one]

if your pc is within 3 to 4 years old its uefi. ubuntu will work fine and grub as the boot loader. pretty much any pc that came after win 8 and after is uefi.

[eviltracy]

I did this, using wsl 2, the windows store ubuntu app, and deoldify.

RTX renders work great.

will do a write up if interested


[Attachment 55634 - Click to enlarge]

[Joey Bagodonuts]

Thanks, eviltracy....

I would love to see your procedure on how to do this!
Most appreciated.

Joe

[Wester de Weerdt]

I did this, using wsl 2, the windows store ubuntu app, and deoldify.

RTX renders work great.

will do a write up if interested


[Attachment 55634 - Click to enlarge]

Could you share the detailed steps to get it working on Windows?

[matt8000]

I've been following this thread with interest. Joey Bagodonuts: If you want to take advantage of the processing speed of a GPU then as far as I know the only way to do that at the moment on your machine would be to dual boot. I eventually found out WSL doesn't currently allow Linux to know what GPU you have. Meanwhile if you were to install Ubuntu inside a VMWare Virtual Machine and set this up, it will work, but again no GPU will be recognised forcing you to use CPU. I did read that it was possible to get Linux to see the GPU if the server version of Ubuntu was installed on specific settings but I had no joy there either. I've successfully set this up in WSL and VMWare though just using the CPU. It took some doing. Are you still requiring help setting it up?

[Joey Bagodonuts]

Answering a PM here....as it makes no sense to share info privately....

I was successful in 3 things:
Dual Booting Win/Ubuntu and running DeOldify within Ubuntu
Running Ubuntu and DeOldify in a Virtual machine within Windows
Running DeOldify in Windows itself through Anaconda (which is a shell in its own right, but can run within Windows).

As was mentioned previously on this thread...unless you are running DeOldify directly in Linux, the GPU is not properly utilized. So running in Windows or in a VM is a waste of time. I did learn a lot however and that is typically worth the price of admission.

The DeOldify model is the best I have seen so far at AI colorization, and will doubt only get better. At this point, the only downside is the re-encoding output quality of the video, which (at least to this novice) is not adjustable. That being said - I look forward to further development and experimentation.

The current version of Vegas Pro has a colorization feature, but it is very weak in comparison to the DeOldify model, IMO.

Kudos and thanks to the DeOldify Team!

[Hoser Rob]

Thanks for that link...

As I alluded to in my OP....you would need to run it in Anaconda, which I believe would be as a "shell" in Windows.
I am looking for how to go about doing that.

I was hoping someone who is familiar with the scripting in that environment could offer assistance.

Anaconda isn't just a 'script'', it's a native Linux program, as is deoldify. You'd need some sort of Linux emulator and those aren't reliable.

[Htomas]

Hello.
I have encountered another problem with starting DeOldify. When I try do run the servive an error appears. I found the instruction :

"Verify Correct Runtime Settings
IMPORTANT

In the "Runtime" menu for the notebook window, select "Change runtime type." Ensure that the following are selected:

Runtime Type = Python 3
Hardware Accelerator = GPU"

But when I open the Runtime menu I got only the Hardware Accelerator. No Python.
How to resolve this problem? My colleague who also use Windows 10 claims that he uses DeOldify wiithout problems.
I tried to start the service as well in Windows as well in Linux Lubuntu (run in virtual machine). In both cases is the same.

[Joey Bagodonuts]

As I mentioned above....I was finally able to get DeOldify working properly in Windows through Anaconda. There are however, 2 issues that make it impractical:
1) when working in Conda (or using a Linux VM in Windows), the GPU is not properly recognized and its memory not properly allocated...therefore, your ability to adjust the Render Factor is extremely limited.
2) whether running natively, or in the Colab online, I have yet to see a way where you can adjust the quality of the output. So you can load in a video file that is 400 MB (for example), but DeOldify will render a completed file at 100 MB (again, just an example....I am not sure of the ratio exactly).

So unless there is a way to solve those 2 issues (there may be others, but I stopped there), there really isn't any reason not to just use the Colab online.

[Oto]

Windows GUI now available here - https://github.com/ColorfulSoft/DeOldify.NET
Only for photos for now!

Features

Support for Windows 7 and above and Linux
The ability to color black-and-white photos using an artistic neural network
Binaries

For Windows: DeOldify.NET.win.exe
For Linux: DeOldify.NET.linux.exe