You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I've tried all models (one by one) on some random videos with random duplicate frames.
To be more accurate I'm now testing with both: REAL FOOTAGE and ANIMATION.
To keep it short:
The solution should NOT CARE if it's Animation or Real Footage, that could be very helpful so it will work with ANY model nicely.
The goal is to scan and recognize and remove duplicated frames a.k.a "same / similar" frames to 1 frame as first process all over the footage and only then start the interpolation process.
I guess that some models supposed to do that but based on my experience it's not working.
At the moment I manually delete all duplicated videos first then render and only then use REAL-Video-Enhancer and it looks nice.
This could happen in older videos or bad captured videos / stream buffer and most known in 2D animations.
The footage may contain sometimes 2, 3 or 4 of the exact or similar frame sometimes.
In the animation industry the most popular terminology for it is "TWOS" or "THREES" and if it's general name is known as "HOLDS" if it's more than two similar frames or in general footage when editing.
🤐 Since I'm an animator I'm dealing with these cases on daily basis, sadly I'm not a programmer so I can't help much beside of sharing my experience, problem and solutions, hopefully it helps.
I wonder if there is a nice option (open source algorithm?) to add such feature to REAL-Video-Enhancer by checking the option to scan/process this and only then start the interpolation as I explained above.
IF it helps:
The only open source software I know that have something similar automatically (not perfect, but not too bad) is: Shutter Encoder but it's algorithm is not new (pre mainstream AI usage).
My guess is that probably by now, there is a pre-trained model or better algorithm in general to do the same thing automatically with more precise on any footage, no matter if it's Animation or Real-Footage.
🤔 If this is possible, it will be a huge feature to have probably under the Advanced Tab as a pre-process option.
The text was updated successfully, but these errors were encountered:
I've tried all models (one by one) on some random videos with random duplicate frames.
To be more accurate I'm now testing with both: REAL FOOTAGE and ANIMATION.
To keep it short:
The solution should NOT CARE if it's Animation or Real Footage, that could be very helpful so it will work with ANY model nicely.
The goal is to scan and recognize and remove duplicated frames a.k.a "same / similar" frames to 1 frame as first process all over the footage and only then start the interpolation process.
I guess that some models supposed to do that but based on my experience it's not working.
At the moment I manually delete all duplicated videos first then render and only then use REAL-Video-Enhancer and it looks nice.
This could happen in older videos or bad captured videos / stream buffer and most known in 2D animations.
The footage may contain sometimes 2, 3 or 4 of the exact or similar frame sometimes.
In the animation industry the most popular terminology for it is "TWOS" or "THREES" and if it's general name is known as "HOLDS" if it's more than two similar frames or in general footage when editing.
🤐 Since I'm an animator I'm dealing with these cases on daily basis, sadly I'm not a programmer so I can't help much beside of sharing my experience, problem and solutions, hopefully it helps.
I wonder if there is a nice option (open source algorithm?) to add such feature to REAL-Video-Enhancer by checking the option to scan/process this and only then start the interpolation as I explained above.
IF it helps:
The only open source software I know that have something similar automatically (not perfect, but not too bad) is: Shutter Encoder but it's algorithm is not new (pre mainstream AI usage).
My guess is that probably by now, there is a pre-trained model or better algorithm in general to do the same thing automatically with more precise on any footage, no matter if it's Animation or Real-Footage.
🤔 If this is possible, it will be a huge feature to have probably under the Advanced Tab as a pre-process option.
The text was updated successfully, but these errors were encountered: