The LipDub platform has its strengths, but can also be challenging if you don’t know all its in’s and outs.
Here’s a quick list of common user errors that can cause delayed or even poor results when using ADVANCED WORKLOW.
The most common mistake is clicking on the “Train” button at not the right time.
Below are a few examples where clicking the “Train” button at the wrong time can actually cause poor or delayed results.
.png?table=block&id=15863dbb-5363-81b9-bb32-e872a85e0350&cache=v2)
Clicking the “Train” button before all videos are finished uploading (including Training footage)
E.g. This means the AI model will be trained on half the footage. And therefore will likely not yield the best possible results
The LipDub interface does provide a warning when videos are not finished uploading.
NOTE: Be sure to also wait until the Training footage is finished uploading as well.
Clicking “Train” button before you double checked all your actor's face detections are properly assigned
If you don’t check the face detections, you run the risk of creating AI models that are trained on incorrect faces.
It is of the upmost importance you double check ALL face detections in the “Label Actors” step to ensure the data is correct before clicking the “Train” button.
How do I double check my actor’s facial trackings?
- Explanation with GIFs: link
- Video Walkthrough: link
Clicking “Train” button because it failed
- If a failure occurs like this one below, please reach out to support@lipdub.ai. Simply clicking Train button again will unlikely solve the problem.
Clicking “Train” button expecting a different result, but not changing the input data
- if the input data does not change, re-training the model will NOT yield a different result. It will be near identical.
