SelfieBarracuda is a human segmentation filter that runs the [MediaPipe Selfie Segmentation Model] on the [Unity Barracuda] neural network inference library.
[MediaPipe Selfie Segmentation Model]: https://google.github.io/mediapipe/solutions/selfie_segmentation.html
[Unity Barracuda]: https://docs.unity3d.com/Packages/com.unity.barracuda@latest
I converted the original model (provided in the tflite format) into ONNX using tflite2tensorflow and a small [Colab notebook].
[Colab notebook]: https://colab.research.google.com/drive/1t40C-cS3q1AYMj8Vpq4pUqdX0XfLr537?usp=sharing
This repository only contains the landscape model (256x144). There is another project called [SelfieSegmentationBarracuda] that contains the general model (256x256), which is better for square/portrait aspect ratios.
[SelfieSegmentationBarracuda]: https://github.com/creativeIKEP/SelfieSegmentationBarracuda