Dumb question, how do you create logs in Video Enhance AI?
This is machine translated from Japanese to English, so please forgive me if the text is difficult to understand.
We will report the updated result of Windows 10.
Processing speed has improved somewhat.
Windows 10 Version 1909
About 42-44 sec./frame
Windows 10 version 2004
About 36-38 sec./frame
However, v1.5.1 could process at approximately 13-15 sec./frame, so it cannot be said to be a very good result.
Moreover, the problem that āUse CPUā cannot be applied was not solved.
The G algorithm crashes me every time in 1.5.2-- Mac Pro 10.15. Known issue?
Luckily I still have 1.5.1 installed and donāt use that one much anyway.
Also any chance yall can add debug param controls? Iām using this for art (as one use case) so I often want MORE artifacts/noiseā¦ guessing it may not be too hard to add a bit more tweakability for those who want it.
If not, no worries. This software changed my life!
Please contact support so we can get your logs and computer data.
You can enable them under Help>Enable Logs restart their program, replicate the issue, then close the program. You can get the logs later by clicking āOpen Log Folderā and then creating a zip of all the files in the folder and sending them to support.
yup cool saw that will capture a crash and send along. thx!
Iāve been testing a lot with the Theia model, I think it has great potential, I would like to make a suggestion on the configurable options.
I think the Reduce Noise range in the Theia model should extend support a wider range, right now even the lowest setting of ā0ā still blurs the surfaces too much, although the edges are kept fine, it would be better if the Noise Reduction of ā0ā actually produces bare minimal noise reduction and keeps a bit more details.
Let me report about major bug Iāve experienced that makes 1.5.2 version unusable at all.
Yesterday Iāve upgraded to it from 1.4.2. Iāve opened and started processing. I always use output to PNGs.
After an hour or so it was an electricity shutdown at the house. I have no UPS. After recovering the power I started the computer hoping to test the new Auto Save/Load feature. Iāve clicked the button but nothing happened.
As always after program crashes Iām checking the last processed frames in the folder because in that case few last of them are becoming corrupted. This time there were no corrupted files but 1500 of the last PNGs had AN ORIGINAL resolution instead of 200% as expected. After discovering this I immediately installed the 1.4.2 version back.
Also the programās UI was 2 or 4x scaled up. But it is just a funny bug. Iāve got a 4K monitor.
this version crashes in macos when using Gaia, is there a way to download the previous version?
Please contact support so we can get your logs and figured out what happened and do further tests with you. Unfortunately we canāt fix issues like this without logs and further testing because we canāt replicate them.
Upgrading from 1.5.1 to 1.5.2 the program wonāt start, it canāt locate MSVCP140_1.dll
Windows 10.0.18363.959
I uninstalled 1.5.2 and reinstalled 1.5.1 since it looks like 1.5.2 has some other issues and I need to finish this project.
Same problem here. Version 1.51 run fine, 1.52 installed successfully but it never loads properly ever since. Task Manager shows that it has started but there is 0% activity.
Fixed by making VEAI admin !
I created logs from version 1.5.1, completely uninstalled (including C:\Users folders and registry keys), then installed and created logs from version 1.5.2.
I sent them along to support.
I will mention here, in looking at the logs myself, is that in version 1.5.1, Err.tzlog is 0 bytes. In version 1.5.2, it contains the following data:
2020-08-30 08:29:31.074057: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX AVX2
2020-08-30 08:29:31.076057: I tensorflow/stream_executor/platform/default/dso_loader.cc:53] Could not dlopen library 'nvcuda.dll'; dlerror: nvcuda.dll not found
2020-08-30 08:29:31.076057: E tensorflow/stream_executor/cuda/cuda_driver.cc:318] failed call to cuInit: UNKNOWN ERROR (303)
2020-08-30 08:29:31.076057: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:169] retrieving CUDA diagnostic information for host: WIN-7**********
2020-08-30 08:29:31.076057: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:176] hostname: WIN-7**********
Not sure if this is relevant to the 4x - 5x CPU slowdown problem, but the first line of the above log may be relevant.
My suspicion is that OpenVINO is not functioning with version 1.5.2 on certain machines.
Continuing the discussion from Video Enhance AI v1.5.2:
When I run the calculation, VEAI crashes.
I go back to 1.5.1
VEAI 1.5.2 crash.txt (113.4 KB)
I donāt know whether this is by design, or a flaw: Cropping a dimension smaller than the original is not allowed. Example, take 1280x720 video and 200% so it is 2560x1440. Then choose custom and set the width to 1000 while leaving the 1440 alone. It bounces the width back to 1280. I was able to get my crop proportionately by making height more than I wanted - using 256%, then cropping width at 1280 so final was 1280x1844.
I was trying VEAI lately, and I noticed, that setting red cropping frame also by mouse/hand and entering numbers in the corresponding fields to fine-tune would be extremely helpful. Also, possibility to see single frame AI-magnified, outside the 30 frames preview feature, just by moving the upper slider - that would be TREMENDOUSLY nice.
I already asked to topaz what you ask, and they are not interested to implement this feature ā¦ so like you, for almost every of my videos, i play with the crop setting until i find something acceptable, playing with %, and heightā¦ as rendering a video with black bars or not in 4/3 or 16-9 take much more time to render than without it, and as my cpu is not a big one, i must do it for each videosā¦ after 30 or 40 clip to crop, Iām very tiredā¦
This is a bit ridiculous, if itās true, considering what ffmpeg can do with video processing/encoding - no problem with cropping setting or stuff like this - I donāt get it. Or did you ask tech support about it? Cause I think devs read this page much more often, so implementation of this function wouldnāt be far off.
For now itās better to crop the video outside the VEAI, and save it only slightly compressed, to not to worsen the quality, and then feed VEAI with it.