Topaz Video AI AB Prototype 4.0.0.p

Topaz Video AI Video AI Beta
1

Hello everyone,

We have a prototype available for testing. This prototype is the new A/B comparison. The main goal of this early prototype is to test the previewing workflow.

This version is in a developmental state, meaning the prototype is very buggy. You can expect a lot of functionality to not work as well as frequent crashes. Export does not work well on the frontend, but you can still expect decent results on the filesystem.

The prototype overhauls how you typically interact with Topaz Video AI. Instead of the typical input, output view that you are used to, now, you will have two views that models can be applied to. With these views you can build and compare different combinations of models to find out which model configuration works best for your original input.

4.0.0.p

Information about the prototype

The main goal of the change is to allow you to make the best decision about what model configurations to use. To accomplish this, we think it’s best for the user to be able to apply different models and see the results side by side.

For example, you can apply Proteus to your A view and Artemis to the B view to see which one has the result you prefer. At any time, you can revert to the original input to compare with the original. This is different from before where you must apply one model configuration at a time and always be comparing to the original.

Details about how to use
The first difference is you must select which view to focus. If you click the left view, A will be highlighted. If you select the right view, B will be highlighted.

When you have a view highlighted, the models/filters you add to the view will be specific for that view. So if you want to see how “Recover Original Detail” works you could give view A & B Enhancement, then give them different “Recover Original Detail” values.

When you have made the changes to the filters you want to see, you can click “Process Next …” to see these two different combinations of filters applied to the input.

When you do this, a progress bar will be applied to the timeline. This is the area of the video where you chose to see the filters be applied. You can watch this section to see how the input has changed. At any time, you may right click a view and click “Reset to Original.”

When you have a combination of filters that you really like (lets assume it’s the filters applied in view A) you may click the A view and click export.

We think this is the best way to be able to find the best mix of models to get the best results possible. Let us know what you think about the new previewing workflow established with this prototype.

10 Likes
2

So what is the main goal of this build we should be looking for? is it the UI changes, look and feel?
or there are improved on the backend models/process and we should expect video output quality/speed improvement?
I understand we need to approach this as a POC, Ignore all bugs, that’s understandable…
However, to evaluate the concept correctly we need more detail what we are looking for… What part should we be focusing on?

If it’s user experience improvements, I guess we should report the user experience concept (not bugs).
If you looking for us to try out backend process I guess that is what we should test to see if the new backend process concept is on the right track.

some guide lines on what to focus on would be great. that way we can focus better on our POC we should conduct.

4 Likes
3

We need more details. :eyes:

1 Like
4

I couldn’t understand how to work out this new previewing system.

5

I updated the post, let me know if that helps!

4 Likes
6

I updated the post with some more details.

2 Likes
7

This is a very nice Idea. I think to make it even more perfect, is if you could offer the same concept as an option with Split view (with the line you drag left-right). then it would be much easier to spot differences. Some might be subtle differences so hard to spot side by side. Similar to Photo AI.

7 Likes
8

I haven’t had a look into it yet but would like to suggest the option to preview a single choosen frame just for the case it is once again a horrible endless loop of motion where I have to fiddle around with the preferences and auto-play all the time to make this motion sickness causing animation to stop!

2 Likes
9

Looking good so far!

Once the A/B functionality has been verified, will there be options to add additional preview panels? Something like the old 4 panel preview?

Thanks

5 Likes
10

CC @ida.topazlabs as well just in case.

@luke.roche , the “rolling cat” video used in the demonstration was filmed by and belongs to me. I give Topaz Labs retroactive permission to use this video for the demonstration of new features.

But I just wanted to bring up this point. A video I submitted as part of bug report was used for demonstration/promotional reasons. It’s just a short video of a cat. And I’m fine with you using it. But what if Topaz Labs used a personal video (E.G. A video containing family members) that the original owner of the video wanted to keep private?
Or if Topaz Labs used a video created by a well known videographer? People might assume the videographer endorses Topaz Labs products by allowing them to use their videos in demonstrations?

There could be consequences as a result of Topaz Labs using videos that they do not own, or is not shared under a open license. So for the future, Topaz Labs should be more careful about which videos they use for the demonstration of Topaz Labs products, ensuring they have permission to use them.

Once again, I’m fine with you using the rolling cat video here.
And I don’t personally blame you @luke.roche . Videos are probably shifted around a lot inside the company and keeping track of who owns the video can be hard without a proper management system.

Also, I just want some clarification. What happens when we share videos with Topaz Labs?

I assume after the file is shared with Topaz Labs, the videos are used for the investigation of the issue we have reported, then probably deleted when they are no longer needed. But that doesn’t seem to be the case.

So what actually happens when we share a video with Topaz Labs? Do the videos get used in the training of AI models? How long are the videos kept by Topaz Labs after we submit them? Will Topaz Labs use our videos for promotional purposes?

6 Likes
11

I thought the use of the dropbox was made especially for privacy purposes. :eyes:

12

Some of the privacy benefits of sharing files with Topaz Labs still exist. For example, only Topaz Labs, and myself should have access to the original video (maybe Dropbox if they want, maybe the public if there is a data leak I don’t know about).

But when people are told they can privately share something with Topaz Labs via the link, they probably assume the video, or parts of the video (like in the demonstration above) aren’t shared publicly. But it’s not explicitly stated this is the case. And that’s where the confusion/concern comes from.

1 Like
13

@20rushtonj , you raised a very impotent topic. in a nutshell , Topaz has to embrace a better data handling process and take in mind all those Data Privacy Act, GDPR, PIPIDA, PDPA, and all it’s flavors.

You are correct that once a user submits personal information, that was well established by the supreme courts that it is not limited to words, that there should be a better data handling / lifecycle process.
Having said that, to soften the blow a bit, let’s remember that Topaz is a startup company that runs on capital funds (I assume) and those type of companies are not yet that matured when it comes to those types of standards and policies. they all (employees) trying to help each other in any way to get the job done, just to make things work and in some cases without putting to much thought on the data source.

What you raised now in your post, may be the beginning of Topaz forming a data handling process / policy, as Topaz are now carries more responsibility on their shoulders.

By the way, it’s an adorable cat :slight_smile:

2 Likes
14

“Trim” and “Crop” don’t work, regardles which “Process Next” get selected the whole clip will be processed. That is frustrating.

15

Hello and thanks for letting us know about this. We take user privacy very seriously and the use of this video in the demo clip was an oversight that won’t happen again.

When setting up a new test environment for development, this video was mistakenly included in the sample video set. We want users to be confident in their privacy when submitting bug reports and media through our Dropbox. We will be reviewing the test videos available to our team and ensuring that user-submitted videos are only used for bug testing/fixing and then removed from our dev machines.

6 Likes
16

Hello, crop and trim are currently disabled for development purposes. But if you don’t want to process the entire clip: to the right(but still inside the same button) of process next there is a drop down arrow. This arrow can be used to select a different amount of time/frames to process. It’s possible your current default is large enough to process the entire clip! See below:
image

Thanks for the feedback!

1 Like
17

There is discussion to have the 4 panel view, I personally think it would be very useful. We will see what we have time to accomplish by the time of rollout, as our main focus is making sure everything works well with this new system! Right now, the main focus is 2 views that work well.

3 Likes
18

As I already said: „Process Next“ does not deliver, regardless which size I select, the whole video gets processed

19

is this an inbuilt feature of topaz video ai or you are using some other tool?

20

A 4 panel view will over challenge the system. Already actual 2 panel view load my system (AMD Ryzen 5800 8 core 32GB with RTX 3060 12GB) to its limits: 80% CPU 17GB, 96% GPU 11GB when I process 4k IRIS, while fps goes down to 50%.