Use FaceMapperFrame.py to annotate a sequence of faces from a directory or video with dots - the coordinates can be saved to a CSV
Right-click and drag on a dot to change size
Click on one of the categories on the right to change a color or reset numbers
Hold CTRL while scrolling on a dot to change the color for all of the dots in that category
CTRL click and drag to select multiple dots (selected dots have dashed outlines)
CTRL + right-click + drag on canvas will rotate all selections around their center
CTRL + right-click + drag on a dot will resize all selections
Right click to remove selections
Press DEL while dots are selected to remove them
CTRL + Click on a dot to select all dots of that part
Double-click on a dot to mark/unmark as guess
Right double click on canvas to mark all selections as guess
- "csv %path/to/csv" will open from a pre-existing file inside of an image directory
- ffmpeg (for video processing)
XMLTransformer turns a CSV (outputted by FaceMapperFrame) or a .pts file into an xml format usable by Dlib
- -g flag for including dots marked as guess in FaceMapper tool
- "-c %path/to/pts" for cropping images based on center of bounding box before training
- Saves copies of the following for each image:
- Image with brightness changed randomly
- Image shifted randomly
This option is highly recommended because coordinates are generated automatically and are added to the training set, making the model better.
Contains some tweaks to Dlib's packaged face landmark detection script.
- Increased file types supported
- "th %threshold_value" will show all faces with a greater confidence than the specified threshold value
- '-s' will save output to a separate folder called 'detected' if that folder exists (currently only functioning if crop is selected)
- '-sm %n" will smooth output by taking best face from n nearest images
- '-sh' will show output in a window
- '-o' for internal testing