gig-aviad's picture
Publish tested MediaPipe face-tracking models and licenses
18cba40 verified
|
Raw History Blame Contribute Delete
1.13 kB
---
license: mit
pipeline_tag: object-detection
tags:
- tflite
- mediapipe
- face-detection
---
# Gigaverse face-tracking models
These three TensorFlow Lite detectors supply face bounding boxes for the Gigaverse web camera worker using MediaPipe Tasks Vision ObjectDetector. They are the conversions tested in gigaverse-app/gigaverse-web-app at commit e73a666cd. The original conversion recipe is unverified.
| File | App algorithm | Upstream |
| --- | --- | --- |
| yunet-256-int8.tflite | yunet-256-int8 | [YuNet](https://github.com/opencv/opencv_zoo/tree/48b10fa505aeb51fbd127fc60a5c660040e823b7/models/face_detection_yunet) |
| yunet-160.tflite | yunet-160-fp32 | YuNet |
| slim-160.tflite | ultraface-slim-160 | [UltraFace](https://github.com/Linzaer/Ultra-Light-Fast-Generic-Face-Detector-1MB/tree/dffdddda9794a50607cba8f318507a28c1c27cab) |
Individual files can be verified against SHA256SUMS. The face-tracking-models.tgz package contains the same files and upstream MIT license notices for use with a package manager. The web app deploys them as separate assets and downloads only the selected detector when needed.