Video Enhance AI v1.6.1

Download: Windows, Mac. Released February 19, 2020.

Fixes

  • Fixed issue where software would crash if output size was an odd value. Outputs for video are now forced to be even sized.
  • Added warning that outputting above UHD can cause issues with the MP4 container
3 Likes

@matt.lathrop

Hi,

in regards to 1.6.1 I noticed the zoom is very large when the program opens, and during the processing the zoom became 2x the size compare to before, it’s too big now, was something regarding the zoom of everything inside the application changed?

Also, I tested the Theia-2 model, the “matrix grid artifacts” has decreased a bit, but it’s still visible, maybe the model needs more tuning?

1 Like

Turn off UI scaling in windows or mac preferences. That will make the UI appear normal. There seems to be a bug here, but we have yet to find a solution other than turning off scaling on the computer.

any improvements to performance? I just updated (yesterday) to 1.6.0, would like to wait before reinstalling [and possibly running into issues]

I noticed under the Theia model, in the “Restore Detail” setting, a higher number setting (100) actually produces LESS details than a lower number setting (0).

Is this intended? if so, I think it is counter-intuitive. if the setting says “Reduce Detail” that would make much more sense that a higher number in the setting would produce LESS detail

2 Likes

I did not load 1.6.0, but noticed that it said that stopping short was fixed, and I’m happy to see that it is. I was a bit worried when I loaded a 6514 frame video and when it ended it was on frame 6510, which is apparently a “bounce back” but the result did have all 6514 frames. Good job!

Now one of my others still not dealt with, which is rotation. Understand that I’m not expecting you to rotate the preview, as that’s a whole separate can of worms. Just look for the rotation meta value, and if it is there write it back in the MP4 output. Viewers, for the most part, understand it and will correctly rotate the video. For now I’m adding the value back myself in post-processing but it seems unnecessary.

I’m seeing some pretty significant differences in color between 1.5.x and 1.6.1. Reds look to be more orange than previous results, and than the original. Is anyone else seeing this?

I’d post examples here, but apparently I’m not allowed to post pictures or links to image sharing sites?

Please don’t post the same post in 2 places in the forum.

Ignoring the color in the actual app is the color shift present when comparing just the input and output video files?

I have snapshots from both input and output files I can share if you’ll tell me how.

I am comparing the generated files, not the view in the app.

Sure.

@MattLathropTL Sir thank you for your work on the new version. I wanted to bring your attention to something. The “forced to be even sized” doesn’t only affects the dimensions of the video but it seems to affect the FPS of the video as well. Ex: I have a cartoon that is 23.976 and it’s 720x480 and after converting it using Gaia CG to 1920x1080 with a bitrate of 41 the result is a video file with FPS of 20.000. I don’t know if this is intentional or is it something you are unaware of. Either way thank you for your good work.

I don’t know why any of the AI programs messes up with the color. If the program would simply ignore the color, it would be much faster? Or at least make an option to disable color processing.

It’s not the fault of the programs but the color space standards that are different between SD/HD and the video players.

anyone could tell me is it not working on RTX3080 ? i just change my display card and keep loading on perpareing in 5mins screen , thanks all guy :smiley:

You’re the one who bought the card at 20.000$ on ebay ? :smiley: (sorry :wink: )

I check with VEAI.exe, still keep loading in 5mins screen , the message show below :

2020-09-19 21:26:07.817349: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX AVX2
2020-09-19 21:26:07.822033: I tensorflow/stream_executor/platform/default/dso_loader.cc:42] Successfully opened dynamic library nvcuda.dll
2020-09-19 21:26:07.843976: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1640] Found device 0 with properties:
name: GeForce RTX 3080 major: 8 minor: 6 memoryClockRate(GHz): 1.74
pciBusID: 0000:07:00.0
2020-09-19 21:26:07.844072: I tensorflow/stream_executor/platform/default/dlopen_checker_stub.cc:25] GPU libraries are statically linked, skip dlopen check.
2020-09-19 21:26:07.844523: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1763] Adding visible gpu devices: 0

How about adding an intelligent frame rate down / up function?

1 Like

Since version 1.6.0, VEAI crashes after maybe 2 videos when I attempt to batch process using MP4.

It seems you’ve replaced the audio stream copy in the new release from ffmpeg to an in-house copy process? However the new version downgrades my Dolby 5.1 audio stream to plain stereo… …at least that’s true for a little under 8 minutes atfter which the audio goes completely dead… this is true for two titles I’ve tried…I sincerely wonder if anyone has bothered to test this properly. Even with a fast graphics card (I’m using an rtx 2080 ti) it can take HOURS to process a video. Please try and imagine how frustrating it is to discover that, after waiting for many hours, the result is defective. Thankfully I know how to fix the audio using ffmpeg otherwise I would be a lot angrier than I am at the moment…