Video AI 3.0.0.7.b

It is for parallelism. Instances = 0 means single threaded, 1 will use 2 threads and so on. Tensorrt models only support single instance and ignore the instances value.

1 Like

So do veai_fi and/or veai_up support it?

Can you turn on detailed logging in preferences, try to run the models again and if they don’t work post the logs. Help-> logging-> gather logs for support

The idea is to have both with a preference to switch between the ascending and descending order.
Same is the case for preview playback, frame number vs time code, most gui settings.

2 Likes

We are aware of it, Apollo is supposed to handle motion blur better can you give it a try.

Unfortunately scaling works better when symmetric so powers of 2 are the best.

Here goes. Lossy progressive input.
I asked VEAI for a 2sec preview. Proteus and Artemis previewed okay. Gaia “model failed” and Theia in both Fidelity and Detail caused “Model Not Found”.
logsForSupport.tar.zip (1.2 KB)

You guys will obviously know better then I do but I do see a “403” error in the logs when trying to download those models. I also see a space in the log file when it is reporting that;
Critical Attempt 0 failed for http://veai-models.topazlabs.com /ghq-v5-ghq-v[V]-gaia-fp16-4x-ov.tz
So, if VEAI is really trying to download that file but puts a space in the filename string, it is then going to try to download the main models URL which will of course return a 403 forbidden.

1 Like

Thank you for letting us know. Can you replace these files in the model directory and let me know if they work. This will be fixed in the next early access and beta versions.

models.zip (7.8 KB)

1 Like

A Contradictory Behavior! Previously, a preview segment would loop continuously. In VEAI 3-7b, the Play/Pause prompt helpfully suggests that it does. But it really doesn’t (and it really should.)

:nerd_face:

An Inconsistent Contradictory Behavior! The Cortradictory behavior above is only true if you Click on the Play/Pause button on the timeline and after the preview’s original rendering process is done. If you simply click on the line for any given preview, it loops correctly, as it had ought to be doing in the first case.*

:thinking:

Thanks. Downloading that models file got Theia working straight away.
Gaia still fails. I can confirm that the file ghq-v5-gaia-fp32-288x288-4x-ox.tz is present where VEAI is looking for it, though I do notice that the logs show the file location with backslashes except for the final which is a forward slash - I of course don’t know whether that is why VEAI is complaining about not being able to read it.

2022-09-17 20-53-02 Thread: 34460 Debug Updating video info {“sar”:1,“framerate”:29.96853645059674,“startNumber”:1,“frames”:60,“progress”:0,“status”:2,“frame”:0,“procStatus”:{“status”:2,“eta”:0,“fps”:0,“message”:" Model Failed",“pass”:1,“error”:“”,“progress”:0,“frame”:0,“priority”:3,“requestPos”:6,“processorIndex”:0}}
2022-09-17 20-53-02 Thread: 34460 Info OUT: 8 2022-09-17 20-53-02 Thread: 41292 Critical Loading default error reading file: “C:\ProgramData\Topaz Labs LLC\Topaz Video AI BETA\models/ghq-v5-gaia-fp32-288x288-4x-ox.tz”

:AIERROR: Model Failed

2022-09-17 20-53-02 Thread: 41292 Critical Unable to create or load model of type ONNX from path: “ghq-v5-gaia-fp32-288x288-4x-ox.tz”

2022-09-17 20-53-02 Thread: 41292 Critical BackendRunner cannot be initialized for model: “ghq-v5-gaia-fp32-288x288-4x-ox.tz”

2022-09-17 20-53-02 Thread: 41292 Critical Model Runner init failed for index: 0

Now that you mention it, I remember seeing that post. Something about his ‘workflow’ and his labors as being a beta tester. Finding things on this discussion software is not easy.

Anyway, glad to see from @suraj that they are planning to give an option. But in the meantime, it would be nice to have it as they even knew it would be best, most recent on top. :blush: :+1:

1 Like

GREAT! Stop and resume works fine! Finally you made it.

One of the “gotcha’s” that comes from putting new previews at the bottom of the list is that pressing the Preview button does not scroll the list down, so the latest preview is hidden.

(It was really better with latest preview at the top.)

Another nicety, I’d like to suggest. (Not really a feature, but this would be very helpful
)

When trying to do an A-B comparison of the image(s) in the previews of the latest one vs. the previous, it would be wonderful if the timelines could be in sync. That would facilitate changing back and forth to see what changes happened by changing the settings. - Would make fine-tuning a lot easier.

3 Likes

Mark in and mark out on the timeline no longer do anything
at all. Moving them around doesn’t even change the timestamp readout.

Although preview still loops and disappears/reappears during processing, when processing is complete it doesn’t play like it used to. Pleased to see that!

Edit: nope, preview continues to play after processing, I guess it depends

Also used the crop feature, nice!

Thanks

I just don’t understand


When I click on “Export,” I don’t see anywhere where it’s showing current output info like frames/sec, and it doesn’t preview the way all the 2.x and previous versions did when it was exporting / processing.

Is that correct? When you do an export it just runs in the background with no previews of the export processing, no info, no frames/sec, no ETA, etc.?

Or am I doing something wrong?

Edit - Yes, I was. I am supposed to click the little number in the lower right hand corner of the screen. That took me a while to figure out
 it didn’t come intuitively for me.

The only reason I can see me using this is for the “Auto” feature of Proteus, and Proteus version 3. While the application seems to be somewhat improved, it’s still extremely confusing to me. That wasn’t the case with 2.x and below. They were intuitive to me. This isn’t.

Can we have an option to move the timeline to the top of the video window, like we had in 2.x?

It’s still not using optimized blocks. I’m only getting 288x288 downloads for models. All versions of VEAI prior to 3.x, for me, downloaded optimized models at 480x384. Is this going to change, or is everything going to stay at 288x288 now?

Proteus 3 is in 2.6.4, and the auto feature as well.