Jump to content

Need some help using Video2x on Google Colab

Mr.Stork
Go to solution Solved by minibois,
1 hour ago, Mr.Stork said:

!python video2x/src/video2x.py -i LOCATION/INPUT_VIDEO_NAME.mp4 -o LOCATION/OUTPUT_VIDEO_NAME.mp4 -d waifu2x_ncnn_vulkan -r 2.0

I think the last 2.0 is the upscale parameter

Can confirm, it is the scaling factor (so i.e. doing "-r 2" on a 1920x1080 makes it 3840x2160).

 

1 hour ago, Mr.Stork said:

There's also a lot of options in GUI Video2x that arent addressed on both Github Page and Colab Notebook Sample, or atleast I coudn't find them. (the option's image below)

I think I have to dive a bit deeper into the documentation, to see if all these options are present in the command line interface.

Waifu2x, which I was introduced by Taran here, upscales images using Machine Learning, Video2x seperates frames from video gives it to Waifu2x which upscales said frames, and video2x stitches them back.

The problem with this is, it's very computationally heavy even for my 1660TI. This is why I want to use Colab.

 

A colab notebook is provided on the Video2x page, I'm not sure how to add a video from my Google Team Drive, or how to change settings or how to export it back. (don't have much space in my Normal Drive thats why I use Team Drive). Here's the example Colab Notebook link by @Felixkruemel from below image, someone please explain the how to add, export and change settings of videos.

 

Also, the free GPUs might not have enough vram to upscale to 1440p or even 1080p, so I need to test it out after I figure out how to do it. Thanks.

 

image.png.0bbfb08a317eb3f3fd81748b019acb14.png

 

Link to comment
Share on other sites

Link to post
Share on other sites

If I'm read the information in the link correctly, you just log in to Google Colab (remote into it?) instal Video2X and all its dependencies and then run the software like normal, it just  outputs to your Google drive.

Here they have some information on how the command line interface works, under "Running Video2X (CLI)": https://github.com/k4yt3x/video2x

 

The listed GPU's should have plenty of VRAM for upscaling, as they all have over 8GB of VRAM (and my 2070 Super with 8GB was able to upscale to 4K just fine, it just takes a while).

"We're all in this together, might as well be friends" Tom, Toonami.

 

mini eLiXiVy: my open source 65% mechanical PCB, a build log, PCB anatomy and discussing open source licenses: https://linustechtips.com/topic/1366493-elixivy-a-65-mechanical-keyboard-build-log-pcb-anatomy-and-how-i-open-sourced-this-project/

 

mini_cardboard: a 4% keyboard build log and how keyboards workhttps://linustechtips.com/topic/1328547-mini_cardboard-a-4-keyboard-build-log-and-how-keyboards-work/

Link to comment
Share on other sites

Link to post
Share on other sites

On 2/26/2021 at 9:23 PM, minibois said:

If I'm read the information in the link correctly, you just log in to Google Colab (remote into it?) instal Video2X and all its dependencies and then run the software like normal, it just outputs to your Google drive.

Here they have some information on how the command line interface works, under "Running Video2X (CLI)": https://github.com/k4yt3x/video2x

 

The listed GPU's should have plenty of VRAM for upscaling, as they all have over 8GB of VRAM (and my 2070 Super with 8GB was able to upscale to 4K just fine, it just takes a while).

EDIT: After doing some experiments, I figured out how to import videos and export them from and to any location of my drive/team drive, but the naming has to be manually done. It'd be nice to get a code that batches all videos in a folder and autoupscales adding a "_upscale.mp4" at the end of name.

 

Note: LOCATION only works inside drive

!python video2x/src/video2x.py -i LOCATION/INPUT_VIDEO_NAME.mp4 -o LOCATION/OUTPUT_VIDEO_NAME.mp4 -d waifu2x_ncnn_vulkan -r 2.0

I think the last 2.0 is the upscale parameter, I'm currently testing it by setting it to 1.5 to make sure, I will post an update. (unless I forget)

Also, Im upscaling a 17 sec 1080p60fps video and set it to -r 1.5 (1.5x), it's ETA is 30 mins and it's just pass1, I can't see it being very useful if it's this slow.

Even if I upload a 30fps video, I still can't upscale any avarage length video like anime which is a 20 min video (on avarage)

And Even though Colab is free, I'd have to get a RDP/VPS to keep the notebook active

 

 

SPOILER CONTAINS MY OLD REPLY, BEFORE I DID ALL THE EXPERMENTS

Spoiler

 

Okay but how do I add/import my own video to upscale

 

In the Github link, this below code is written, nnd I coudn't find this line anywhere on the Colab Notebook




python video2x.py -i sample-input.mp4 -o sample-output.mp4 -r 2 -d waifu2x_caffe -- --tta 1

Where is the "sample-input.mp4" located? And I assume the output is same as the input?

What does "-r -d" mean? I assume "2" means, 2x upscale, -i is input, and -o is output?
I understand tta 1 means tta true, and tta 0 would be false.

 

I did however, sucessfully upscaled the smple video in colab




!wget -O video.mp4 http://ffg.gg/s



!pip install -U PyYAML
!python video2x/src/video2x.py -i /content/video.mp4 -o video_4x.mp4 -d waifu2x_ncnn_vulkan -r 4.0

But I'm mainly concerned about adding my own video location, cuz it the code in the Colab Sample is set to download a sample video from http://ffg.gg/s as video.mp4. (above code taken from notebook)

Then it saves it as "video_4x_2.mp4" in my drive

 

 

 

There's also a lot of options in GUI Video2x that arent addressed on both Github Page and Colab Notebook Sample, or atleast I coudn't find them. (the option's image below)

image.png.f560fadd8b7a63dd0b83374959112db1.png

Edited by Mr.Stork
I experimented a bit, and posted my update
Link to comment
Share on other sites

Link to post
Share on other sites

1 hour ago, Mr.Stork said:

!python video2x/src/video2x.py -i LOCATION/INPUT_VIDEO_NAME.mp4 -o LOCATION/OUTPUT_VIDEO_NAME.mp4 -d waifu2x_ncnn_vulkan -r 2.0

I think the last 2.0 is the upscale parameter

Can confirm, it is the scaling factor (so i.e. doing "-r 2" on a 1920x1080 makes it 3840x2160).

 

1 hour ago, Mr.Stork said:

There's also a lot of options in GUI Video2x that arent addressed on both Github Page and Colab Notebook Sample, or atleast I coudn't find them. (the option's image below)

I think I have to dive a bit deeper into the documentation, to see if all these options are present in the command line interface.

"We're all in this together, might as well be friends" Tom, Toonami.

 

mini eLiXiVy: my open source 65% mechanical PCB, a build log, PCB anatomy and discussing open source licenses: https://linustechtips.com/topic/1366493-elixivy-a-65-mechanical-keyboard-build-log-pcb-anatomy-and-how-i-open-sourced-this-project/

 

mini_cardboard: a 4% keyboard build log and how keyboards workhttps://linustechtips.com/topic/1328547-mini_cardboard-a-4-keyboard-build-log-and-how-keyboards-work/

Link to comment
Share on other sites

Link to post
Share on other sites

8 minutes ago, minibois said:

I think I have to dive a bit deeper into the documentation, to see if all these options are present in the command line interface.

Cool, do you know what happes if I try to upscale a .mkv or anything other than .mp4?

Also I think as it's using AI, woudn't it be better if it used the TPU instead of the GPU, like how Nvidia's DLSS works?

Woudn't it be great to upscale videos using DLSS, in realtime or at least way faster than this Colab method

Edit: Apprantly it is possible via something called NVIDIA NGX, problem is I only have a Cuda GPU

Link to comment
Share on other sites

Link to post
Share on other sites

9 minutes ago, Mr.Stork said:

Cool, do you know what happes if I try to upscale a .mkv or anything other than .mp4?

The ffmpeg backbone of Video2X probably handles that just fine. I don't expect you'd get worse results or whatever, but the file size might differ (not sure if 'differ' is positive or negative in this context).

10 minutes ago, Mr.Stork said:

Also I think as it's using AI, woudn't it be better if it used the TPU instead of the GPU, like how Nvidia's DLSS works?

Woudn't it be great to upscale videos using DLSS, in realtime or at least way faster than this Colab method

Edit: Apprantly it is possible via something called NVIDIA NGX, problem is I only have a Cuda GPU

I know Video2X runs on CUDA. I have a Turing Nvidia card, so I could take a look around to see if there would be any improvement.

"We're all in this together, might as well be friends" Tom, Toonami.

 

mini eLiXiVy: my open source 65% mechanical PCB, a build log, PCB anatomy and discussing open source licenses: https://linustechtips.com/topic/1366493-elixivy-a-65-mechanical-keyboard-build-log-pcb-anatomy-and-how-i-open-sourced-this-project/

 

mini_cardboard: a 4% keyboard build log and how keyboards workhttps://linustechtips.com/topic/1328547-mini_cardboard-a-4-keyboard-build-log-and-how-keyboards-work/

Link to comment
Share on other sites

Link to post
Share on other sites

15 minutes ago, minibois said:

The ffmpeg backbone of Video2X probably handles that just fine. I don't expect you'd get worse results or whatever, but the file size might differ (not sure if 'differ' is positive or negative in this context).

I know Video2X runs on CUDA. I have a Turing Nvidia card, so I could take a look around to see if there would be any improvement.

Which one? I have a 1660Ti, and it still chugs

Link to comment
Share on other sites

Link to post
Share on other sites

7 minutes ago, Mr.Stork said:

Which one? I have a 1660Ti, and it still chugs

2070 Super.

Can't say it's super fast, but still muuuuch faster than not having a CUDA compatible GPU (running on a CPU is a lot slower).

 

You did give me some more inspiration to work some more on Video2X, after mostly doing Waifu2X stuff.

"We're all in this together, might as well be friends" Tom, Toonami.

 

mini eLiXiVy: my open source 65% mechanical PCB, a build log, PCB anatomy and discussing open source licenses: https://linustechtips.com/topic/1366493-elixivy-a-65-mechanical-keyboard-build-log-pcb-anatomy-and-how-i-open-sourced-this-project/

 

mini_cardboard: a 4% keyboard build log and how keyboards workhttps://linustechtips.com/topic/1328547-mini_cardboard-a-4-keyboard-build-log-and-how-keyboards-work/

Link to comment
Share on other sites

Link to post
Share on other sites

32 minutes ago, Mr.Stork said:

Also I think as it's using AI, woudn't it be better if it used the TPU instead of the GPU, like how Nvidia's DLSS works?

If you manage to get your code working on the TPU, then sure. Also, Nvidia's DLSS doesn't use "TPUs", it uses the tensor cores built into the RTX GPUs (which are dumbed down "TPU-units") along with the regular stream processors.

 

33 minutes ago, Mr.Stork said:

Woudn't it be great to upscale videos using DLSS, in realtime or at least way faster than this Colab method

There are many models that currently try to do such stuff in real time, such as:

(this one has full source and data available)

and this other one from facebook:

(not sure if they have released the code yet)

 

The ones you're using just aren't meant for speed, but rather quality.

 

38 minutes ago, Mr.Stork said:

Edit: Apprantly it is possible via something called NVIDIA NGX, problem is I only have a Cuda GPU

You need to be a developer and have a company email to register for their beta phase anyway.

 

 

You got me curious about that thing, gonna try to run it on my 2060 Super.

FX6300 @ 4.2GHz | Gigabyte GA-78LMT-USB3 R2 | Hyper 212x | 3x 8GB + 1x 4GB @ 1600MHz | Gigabyte 2060 Super | Corsair CX650M | LG 43UK6520PSA
ASUS X550LN | i5 4210u | 12GB
Lenovo N23 Yoga

Link to comment
Share on other sites

Link to post
Share on other sites

There's a issue major with Colab tho, the "colab work" space is limited to 70gb, while upscaling a 20mb video to 90mb, it used nearly 50gb of space, so I can't imagin using the free version to upscale bigger videos, and also as it extracts lossless .pngs it takes much more, this is why I like to use .jpg on my GUI video2x, see if you can find out how to change the extension

12 minutes ago, minibois said:

but still muuuuch faster than not having a CUDA compatible GPU

 

wait so the GPUs on Colab have Cuda right? or they just gpus wih no cuda?

 

12 minutes ago, minibois said:

You did give me some more inspiration to work some more on Video2X, after mostly doing Waifu2X stuff.

I'm Glad :), let me know if you find something interesting, im always up for these types of ai stuff to improve my workflow. I do graphics and animation stuff, and these stuff helps a lot, if I ever get a low res video or image.

11 minutes ago, igormp said:

There are many models that currently try to do such stuff in real time, The ones you're using just aren't meant for speed, but rather quality.

That's great! I actually want something that would enhance my anime viewing experience to 4k, and it doesnt need to be real time, but a resonable amount of time, like for a 20 min video, 10-5 mins of encode/upscale

 

14 minutes ago, igormp said:

You got me curious about that thing, gonna try to run it on my 2060 Super.

Sure but tensor cores doesnt help much with video2x, it mainly uses CUDA, so wont be much useful. Only GPUs that would be best, I think, are the Quadros. Also glad I could make u curious! 🙂

12 minutes ago, igormp said:

(this one has full source and data available)

Will check it out, to see if I can use it for my content viewing experience

Link to comment
Share on other sites

Link to post
Share on other sites

13 minutes ago, Mr.Stork said:

wait so the GPUs on Colab have Cuda right?

They do, just remember to enable the GPU kernel.

 

14 minutes ago, Mr.Stork said:

Sure but tensor cores doesnt help much with video2x, it mainly uses CUDA, so wont be much useful

They would, since it uses neural networks and matrix FMA on tensor cores is way faster than your regular CUDA core. However, none of the backends for video2x support those, there's even an open issue on ncnn and waifu2x-ncnn-vulkan repos: 
https://github.com/Tencent/ncnn/issues/1992
https://github.com/nihui/waifu2x-ncnn-vulkan/issues/93

 

16 minutes ago, Mr.Stork said:

Only GPUs that would be best, I think, are the Quadros

Other than having more vram (allowing for higher batch sizes), it wouldn't make a difference.

17 minutes ago, Mr.Stork said:

Will check it out, to see if I can use it for my content viewing experience

Be aware that it's not meant for a final user usage, it's research code, and it's not really the best when it comes to code quality.

 

 

17 minutes ago, Mr.Stork said:

There's a issue major with Colab tho, the "colab work" space is limited to 70gb, while upscaling a 20mb video to 90mb, it used nearly 50gb of space,

That's something that I've noticed as well, you need lots of space available for all of the raw frames. Another issue with video2x is that it's whoever wrote it is not really that good with ffmpeg (which is understandable, ffmpeg is basically black magic). The step to get the total amount of frames was taking too long because they forgot to add a switch to use all of the threads available.

FX6300 @ 4.2GHz | Gigabyte GA-78LMT-USB3 R2 | Hyper 212x | 3x 8GB + 1x 4GB @ 1600MHz | Gigabyte 2060 Super | Corsair CX650M | LG 43UK6520PSA
ASUS X550LN | i5 4210u | 12GB
Lenovo N23 Yoga

Link to comment
Share on other sites

Link to post
Share on other sites

5 hours ago, Mr.Stork said:

There's also a lot of options in GUI Video2x that arent addressed on both Github Page and Colab Notebook Sample, or atleast I coudn't find them. (the option's image below)

I check out the program a bit more this evening and I saw in the "\video2x-master\bin" folder, there is a file "video2x.json", which has some of the options I see in the GUI.

 

Currently I am upscaling a 320x240 video to 1920x1080 (6x upscale).

I am honestly not expecting much of a good result out of it, but just wanted to try it out.

this 2 minute 12 second video is taking about 2 hours and 40 minutes, so it definitely takes a good while

"We're all in this together, might as well be friends" Tom, Toonami.

 

mini eLiXiVy: my open source 65% mechanical PCB, a build log, PCB anatomy and discussing open source licenses: https://linustechtips.com/topic/1366493-elixivy-a-65-mechanical-keyboard-build-log-pcb-anatomy-and-how-i-open-sourced-this-project/

 

mini_cardboard: a 4% keyboard build log and how keyboards workhttps://linustechtips.com/topic/1328547-mini_cardboard-a-4-keyboard-build-log-and-how-keyboards-work/

Link to comment
Share on other sites

Link to post
Share on other sites

Overall Colab is great, I use it almost reguraly to transfer files and download from and to my drive, there's a ton of potential. The BIGGEST issue with colab is it disconnects if I shut down the pc/close browser, well I guess it's understandabe. Only workaround is to get a cheap RDP or a server to run browser 24/7 at home like rasberry pi.

 

14 hours ago, igormp said:

However, none of the backends for video2x support those, there's even an open issue on ncnn and waifu2x-ncnn-vulkan repos:
https://github.com/Tencent/ncnn/issues/1992
https://github.com/nihui/waifu2x-ncnn-vulkan/issues/93

That is what I meant actually, I thought it'd be better with those tensor cores as they are specifically designed for AI use. And I've also seen open issues before about the same thing, but there's no sign of that being implemented. Those threads are almost a year old.

 

14 hours ago, igormp said:

Be aware that it's not meant for a final user usage, it's research code, and it's not really the best when it comes to code quality.

Im not much of a coder, so it seems reading and using the code will be hard

14 hours ago, igormp said:

That's something that I've noticed as well, you need lots of space available for all of the raw frames

Which can be fixed by paying for Colab Premium... But not worth it, imo.

 

14 hours ago, igormp said:

Another issue with video2x is that it's whoever wrote it is not really that good with ffmpeg (which is understandable, ffmpeg is basically black magic). The step to get the total amount of frames was taking too long because they forgot to add a switch to use all of the threads available.

Do you know how to add the "switch" to the colab code? The windows version doesn't take as much time to extract frames either.

 

11 hours ago, minibois said:

I am honestly not expecting much of a good result out of it, but just wanted to try it out.

Waifu2x itself is really good, but as it's doing individual frames, there could be issues with continuety. But it's good, unless u pixed peep.

 

11 hours ago, minibois said:

this 2 minute 12 second video is taking about 2 hours and 40 minutes, so it definitely takes a good while

Lol my 17sec video took around 20-30 mins. So did you go through with it or left it? if you didnt how was it?

Link to comment
Share on other sites

Link to post
Share on other sites

3 hours ago, Mr.Stork said:

Lol my 17sec video took around 20-30 mins. So did you go through with it or left it? if you didnt how was it?

I'm kinda salty, because yes I did let it run, but at the very end it gave an error and the video did not save at all.

 

The error? I added one extra letter to the output name which caused it to give an "folder doesn't exist!" error.

Just imagine I wanted it to save as "D:/Videos/Upscaling/output.mp4", but I accidentally set the path as "D:/Videos/Upscalingg/output.mp4"

From a perspective of someone who programs, I feel like this should be the first thing it checks for, at the very least before using up 3 hours of computer time 😛

 

Anyways, I then set my two remaining braincells to work, when I realized "oh wait, this video is basically just a 0.33 second animation, repeated", so I just cut out a part out of the video, upscaled that (to 3840x2880, a 12x upscale), which took a couple minutes and stitched back together using a video editing program.

Of course that didn't go super well too, because that 3840x2880 video has letter- and pillarboxes. After Googling for a bit, turns out the video editing software only support videos with a horizontal resolution of up to 3840 and a vertical resolution of 2160 in the free version I'm using.

 

This was the result I ended up with:

 

Comparison to the original:

Spoiler

 

 

Currently working on another upscale, which I think will go a lot better.

What I did experience though, my video editing program of choice uses quite a bit of VRAM, so I couldn't have it open (just idle after exporting a video) and upscale at the same time, as that resulted in a low memory error from Video2X.

5 hours ago, Mr.Stork said:

Waifu2x itself is really good, but as it's doing individual frames, there could be issues with continuety. But it's good, unless u pixed peep.

My main issue with it is usually the fuzzy lines, which I don't really know how to fix that.

"We're all in this together, might as well be friends" Tom, Toonami.

 

mini eLiXiVy: my open source 65% mechanical PCB, a build log, PCB anatomy and discussing open source licenses: https://linustechtips.com/topic/1366493-elixivy-a-65-mechanical-keyboard-build-log-pcb-anatomy-and-how-i-open-sourced-this-project/

 

mini_cardboard: a 4% keyboard build log and how keyboards workhttps://linustechtips.com/topic/1328547-mini_cardboard-a-4-keyboard-build-log-and-how-keyboards-work/

Link to comment
Share on other sites

Link to post
Share on other sites

59 minutes ago, minibois said:

Just imagine I wanted it to save as "D:/Videos/Upscaling/output.mp4", but I accidentally set the path as "D:/Videos/Upscalingg/output.mp4"

Why woundn't the code the make a folder named 'Upscalingg", that would and does help out in these situations

 

1 hour ago, minibois said:

Comparison to the original:

looks wayy better, I'd do this with every 720p video if it werent for the ludacris time needed

 

1 hour ago, minibois said:

My main issue with it is usually the fuzzy lines, which I don't really know how to fix that.

I think it's because it cuts the video cuz it dont have enough vram to upscale the whole frame at once, DIAN app does that, (DIAN app is an AI slowmo/fps adding app)

Link to comment
Share on other sites

Link to post
Share on other sites

7 hours ago, Mr.Stork said:

That is what I meant actually, I thought it'd be better with those tensor cores as they are specifically designed for AI use. And I've also seen open issues before about the same thing, but there's no sign of that being implemented. Those threads are almost a year old.

I'm not familiar with the frameworks they're using. If it were tensorflow or keras, I could add mixed-precision FP16 easily by myself.

 

7 hours ago, Mr.Stork said:

The BIGGEST issue with colab is it disconnects if I shut down the pc/close browser, well I guess it's understandabe. Only workaround is to get a cheap RDP or a server to run browser 24/7 at home like rasberry pi.

It'll disconnect after 12 hours anyway. It's meant to be used for studies, not as a ML server.

 

7 hours ago, Mr.Stork said:

Which can be fixed by paying for Colab Premium... But not worth it, imo.

 

I ran locally since the hardware I have is better than the free one in colab :old-tongue:

 

7 hours ago, Mr.Stork said:

Do you know how to add the "switch" to the colab code? The windows version doesn't take as much time to extract frames either.

No idea, here's the patch in case you want to apply it:

diff --git a/src/wrappers/ffmpeg.py b/src/wrappers/ffmpeg.py
index b7dab3e..972c46b 100755
--- a/src/wrappers/ffmpeg.py
+++ b/src/wrappers/ffmpeg.py
@@ -93,6 +93,8 @@ class Ffmpeg:
 
         execute = [
             self.ffmpeg_probe_binary,
+            "-threads",
+            "16",
             "-v",
             "quiet",
             "-count_frames",

 

18 minutes ago, Mr.Stork said:

Why woundn't the code the make a folder named 'Upscalingg", that would and does help out in these situations

Because they didn't implement it, feel free to make a PR lol

 

 

FX6300 @ 4.2GHz | Gigabyte GA-78LMT-USB3 R2 | Hyper 212x | 3x 8GB + 1x 4GB @ 1600MHz | Gigabyte 2060 Super | Corsair CX650M | LG 43UK6520PSA
ASUS X550LN | i5 4210u | 12GB
Lenovo N23 Yoga

Link to comment
Share on other sites

Link to post
Share on other sites

13 hours ago, igormp said:

It'll disconnect after 12 hours anyway. It's meant to be used for studies, not as a ML server.

I found a simple script to hopefully that, havent tested it though. Surely it won't disconnect while a cell is running.

 

13 hours ago, igormp said:

No idea, here's the patch in case you want to apply it:

And where would I add it in the code?

 

13 hours ago, igormp said:

Because they didn't implement it, feel free to make a PR lol

Why bother, No one responds anyway

Link to comment
Share on other sites

Link to post
Share on other sites

6 hours ago, Mr.Stork said:

I found a simple script to hopefully that, havent tested it though. Surely it won't disconnect while a cell is running.

 

It will, the VMs that your colab session runs on have a max lifespan of 12 hours, so it'll die no matter what you're doing. The service is free for a reason, don't abuse it.

6 hours ago, Mr.Stork said:

And where would I add it in the code?

It's right there in the diff, src/wrappers/ffmpeg.py line 93

 

6 hours ago, Mr.Stork said:

Why bother, No one responds anyway

They do seem to respond and merge PRs rather quickly.

FX6300 @ 4.2GHz | Gigabyte GA-78LMT-USB3 R2 | Hyper 212x | 3x 8GB + 1x 4GB @ 1600MHz | Gigabyte 2060 Super | Corsair CX650M | LG 43UK6520PSA
ASUS X550LN | i5 4210u | 12GB
Lenovo N23 Yoga

Link to comment
Share on other sites

Link to post
Share on other sites

20 hours ago, Mr.Stork said:

Why woundn't the code the make a folder named 'Upscalingg", that would and does help out in these situations

¯\_(ツ)_/¯ their reasoning is beyond me..

I just tried - obviously with a small video - to output to a non-existing folder again and it gave me the same error.

Command:

python video2x.py -m gpu -r 2 -i D:\Libraries\Videos\Upscaling\test.mp4 -o D:\Libraries\Videos\Upscaling\TEMP\output.mp4

the folder where test.mp4 is located in exists, but the folder "TEMP" in the output does not exist.

video2x cuts the video in parts, upscales it, stitches it back together and only then it checks if the folder it needs to output into actually exists and throw this error:

D:\Libraries\Videos\Upscaling\TEMP\output.mp4: No such file or directory
[!] ERROR: An exception has occurred

I do have to mention I am not on the latest release, so maybe it's fixed already, so I have to try this with the newest version of the software still.

"We're all in this together, might as well be friends" Tom, Toonami.

 

mini eLiXiVy: my open source 65% mechanical PCB, a build log, PCB anatomy and discussing open source licenses: https://linustechtips.com/topic/1366493-elixivy-a-65-mechanical-keyboard-build-log-pcb-anatomy-and-how-i-open-sourced-this-project/

 

mini_cardboard: a 4% keyboard build log and how keyboards workhttps://linustechtips.com/topic/1328547-mini_cardboard-a-4-keyboard-build-log-and-how-keyboards-work/

Link to comment
Share on other sites

Link to post
Share on other sites

16 minutes ago, minibois said:

so maybe it's fixed already, so I have to try this with the newest version of the software still.

It's not, I'm on the master branch.

FX6300 @ 4.2GHz | Gigabyte GA-78LMT-USB3 R2 | Hyper 212x | 3x 8GB + 1x 4GB @ 1600MHz | Gigabyte 2060 Super | Corsair CX650M | LG 43UK6520PSA
ASUS X550LN | i5 4210u | 12GB
Lenovo N23 Yoga

Link to comment
Share on other sites

Link to post
Share on other sites

4 hours ago, igormp said:

It's right there in the diff, src/wrappers/ffmpeg.py line 93

cool thanks

 

4 hours ago, igormp said:

They do seem to respond and merge PRs rather quickly.

Im lazy as is lol

 

4 hours ago, minibois said:

¯\_(ツ)_/¯ their reasoning is beyond me..

haha

 

Anyway, thank you both @igormpand @miniboisfor helping me out a lot, if you guys wanna talk a bit about this topic, I can keep the thread open, if not I'll close it

 

Edit: I just had an idea by watching @minibois's Foldin@Home badge, can we use Colab for Folding@Home? It is for the greater good right?

Link to comment
Share on other sites

Link to post
Share on other sites

3 hours ago, Mr.Stork said:

Edit: I just had an idea by watching @minibois's Foldin@Home badge, can we use Colab for Folding@Home? It is for the greater good right?

That idea reminds me of this: 

Which uses Google Cloud Engine, not sure how similar that is to Google Colab.

I don't know what Google's stance would be towards F@H on Google Colab too.

 

Anyways, on the topic of upscaling, I upscaled another thing yesterday and have been working on another upscale at the moment.

What I did yesterday: 

Upscaled this from 480p to 2880x2160 (4K with pillarboxes).

Working on a 720p upscale at the moment, which I hope will turn out decent

"We're all in this together, might as well be friends" Tom, Toonami.

 

mini eLiXiVy: my open source 65% mechanical PCB, a build log, PCB anatomy and discussing open source licenses: https://linustechtips.com/topic/1366493-elixivy-a-65-mechanical-keyboard-build-log-pcb-anatomy-and-how-i-open-sourced-this-project/

 

mini_cardboard: a 4% keyboard build log and how keyboards workhttps://linustechtips.com/topic/1328547-mini_cardboard-a-4-keyboard-build-log-and-how-keyboards-work/

Link to comment
Share on other sites

Link to post
Share on other sites

4 hours ago, Mr.Stork said:

Edit: I just had an idea by watching @minibois's Foldin@Home badge, can we use Colab for Folding@Home?

No

4 hours ago, Mr.Stork said:

It is for the greater good right?

That's an asshole move IMO. Colab instances are meant to allow people without resources to study and try out stuff for free, doing so is basically abusing the system. If you want to do it for the greater good, then pay for it.

 

23 minutes ago, minibois said:

Which uses Google Cloud Engine, not sure how similar that is to Google Colab.

GCE is just a barebones cloud VM. You can use Colab as a front-end to any jupyter server instead of relying on their free machine, even on your own PC.

FX6300 @ 4.2GHz | Gigabyte GA-78LMT-USB3 R2 | Hyper 212x | 3x 8GB + 1x 4GB @ 1600MHz | Gigabyte 2060 Super | Corsair CX650M | LG 43UK6520PSA
ASUS X550LN | i5 4210u | 12GB
Lenovo N23 Yoga

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×