Custom Exports - MachineEditor/MachineVideoEditor GitHub Wiki
Machine Video Editor allows you to set up your own custom exports, in addition to the ability to use standard DFL ones (like Whole Face, Full Face, etc).
To be able to use your custom export faces with DFL, make sure to use the MVE-DFL fork.
Important info:
- Custom Export settings are saved in project database files (meaning you will need to set it up again for other projects)
- Custom Exports use 2D Alignment Points - the Frames from which you Export will not work with Heads applied to them (Heads use 3D)
- when using custom exports for SRC and DST, the export settings need to be exactly the same for both sets
- if you have two different sets (one for training XSeg, and one with your custom export), make sure both sets are extracted from the same parent frames
First, either open your Faces and Import face data
, or open your Frames that had Faces set to them, and enter the Alignment Editor (right click > Edit Alignments
).
You will be presented with an interface like in the picture below (make sure that Landmarks is selected):
On the left side is the interface for editing alignments, while on the right you will see Face Extraction Preview and options for Custom Extracts.
Marked with a red box, from left to right the options are:
- plus sign - new custom extract
- cog wheel - edit the selected custom extract
- x mark - delete the selected custom extract
Clicking on the plus sign button will open an interface like in the picture below:
Important notice:
You should always do the following:
- disable
Force Align
its use is depreciated
- disable
Ignore Non Visible
it is an experimental feature that is not consistent
- enable
Use Custom Landmarks
- make sure Landmarks 18-49 and 55 are enabled, and 50-54 and 56-68 are disabled
On the top is the field where you can name your custom export.
Padding is basically zoom level - lower values will zoom in more
Vertical Percent Offset moves the center of the picture up or down
Horizontal Coordinate moves the center of the picture left or right
Vertical Coordinate is only used if you enable the checkmark for Use Vertical Output
, additionally moves the center of the picture up or down
Hints
- you can have many custom exports defined, so naming them in a good way can save you time in the future
- changing options here will immediately show you what happens in the Preview window
- when creating new exports, check the preview of several different angle pictures to make sure you are not cutting out facial features you want included
Once you are satisfied with the selected options and how the preview looks, click Save
or Save As
.
Once you have set up your custom export(s), follow the steps from this wiki article, except in the step 5 - select Custom.
By selecting Custom, you will be presented with a new dropdown menu from which you select which of the saved custom exports you want to use.
It will look like in the picture below:
You can train XSeg masks on your regular Face images (WF, Head, etc), apply to those Face images, and then copy the masks to your custom-exported images.
The procedure for that first starts with you opening your Faces (WF, Head, etc) that have XSeg masks applied to them and then:
- open Detection Management from the side menu
- set the directory where your custom-exported pictures are
- enable
Recalculate face data
- click
Copy embedded data
(and wait while it does its thing)