Face Data
A macOS application used to auto-annotate landmarks from a video. Those landmarks can further be used as training data for Generative Adversarial Networks (GANs).
Getting Started
Installing
You can either download the binary file from Rease or build the source code using Xcode.
Use
| Description | |
|---|---|
| Video Path | Path to the video file, currently only support .mp4 files. Use Select File to generate path using a file browsing panel. |
| Output Path | Path to the output directory, this app will create origin and landmarks two sub-directories. Use Select Folder to generate path using a file browsing panel. |
| Start Second | An integer value indicating from which second to start capturing frames from the video, default is 0 (from the beginning) |
| End Second | This app would not extract frames after this second. Default is the duration of the video. |
| # of Frames | Integer value of how many frames you want to generate. Default is 100 frames. |
| Start | Start the process. |
| Cancel | Stop the process. |
Output
- Two sub-directories
originandlandmarkwill be created in the specified output directory. origincontains the original frames extracted from the video, with file name:img001.png.landmarkcontains the landmark image drawn based on the corresponding frame inorigin, with file name:img001lm.png.- If there is no face detected in one original frame, the corresponding file name in
landmarkisno_face_img001lm.png.
Output Images Processing
You will probably want to process the generated images to fit the size restriction for you GANs model. You can refer the Python script crop.py.
Built With
- Apple Vision Library - Easy to reproduce the landmarks in iOS devices
- Apple AV Foundation - Also use lower level image format (
CGImage) to make codes portable to Cocoa Touch