lywcn/mediapipe-models

mediapipe-models

mediapipe-models

Google Mediapipe public TFLite models implemented using Tensorflow-keras (https://mediapipe.dev)

Requirements

• Tensorflow - 1.13.1

Implementation process

1. Hand Tracking

   • Related links

     • Google AI blog post: On device real time hand tracking
     • Github: Mediapipe: Hand tracking

   • Hand Landmark 2D,3D

     • Model Architecture (OK)
     • Set Pretrained weights (OK)
     • Convert TFLite model for 4 channels input (OK)

   • Palm(Hand) Detection

     • Model Architecture (OK)
     • Set Pretrained weights (OK)
     • Convert TFLite model for 4 channels input (OK)

2. Face Detection

   • Related links

     • Paper: "BlazeFace: Sub-millisecond Neural Face Detection on Mobile GPUs"
     • Github: Mediapipe: Face detection

   • Face Detection

     • Model Architecture (OK)
     • Set Pretrained weights (TODO)
     • Convert TFLite model for 4 channels input (TODO)

3. Hair Segmentation

   • Related links

     • Paper: "Real-time Hair segmentation and recoloring on Mobile GPUs"
     • Github: Mediapipe: Hair segmentation

   • Hair Segmentation

     • Model Architecture (TODO)
     • Set Pretrained weights (TODO)
     • Convert TFLite model for 4 channels input (TODO)