Skip to content
Email Support
  • There are no suggestions because the search field is empty.

Training Data

Learn how to select Training Data - if needed

 

What is Training Data?

Training Data is additional video of the character you want to vub. You can upload this to DeepEditor to help improve image quality - if needed.

When do I need Training Data?

If the character you are vubbing has little to no dialogue in your Source media, then:

  • DeepEditor may not have seen enough of the mouth interior to be able to render it accurately;

  • DeepEditor may not be able to render a sufficient range of expressions desired for the new dialogue.

If the character you are vubbing has a complex appearance (variable head pose, significant facial hair or variable lighting) then it is possible that the Source media does not provide sufficient information to DeepEditor..

In these cases, uploading some Training Data of your character will help the image quality.

If the character you are vubbing does have a little dialogue (a second or two) in the Source media where the head pose and lighting conditions are a close match to the section of the shot you want to vub, then you should not need to upload Training Data. 

 

What makes good Training Data?

The most effective Training Data is as similar as possible to the section of your Source media that you wish to vub. Try to meet the following criteria:

  • You only need a few seconds of the person talking. The maximum total duration (if you have multiple clips) is 5000 frames but you should not need anywhere near this amount.
  • The person's head pose is very similar to the section of the Source media you wish to vub.
  • The lighting is very similar to the section of the Source media you wish to vub.
  • There are no black frames.
  • Each Training Data clip features the same actor you are vubbing
  • The person's face is clear and unobstructed

 

How do I create Training Data media?

You would create your Training Data media typically using one of two methods:

  • for Draft quality vubbing
    • by exporting clips from your offline editing software e.g. Premiere Pro
    • if using Avid Media Composer, you cannot use the DeepEditor Avid Extension - which is built for fast iteration using minimal data.
  • for Final quality vubbing
    • by extracting the exact frames you need from your camera original media. This will typically performed by a facility holding the camera original media, or - if you you have access to the camera original media - by using sofware such as DaVinci Resolve.

Before exporting and uploading your Training Data, check that your media meets the file format and codec requirements for DeepEditor.