search

liujianee / SKEPXEL-Skeleton_Pixels_for_Action_Recognition

0 stars 0 forks source link

readme

Skepxels: Spatio-temporal Image Representation of Human Skeleton Joints for Action Recognition

This repository contains the implementation of "Skepxels: Spatio-temporal Image Representation of Human Skeleton Joints for Action Recognition"

Environment

Usage

ENV SETUP

  1. If you are using conda, you can create a virtual environment by conda env create -f environment.yml

  2. Download and config facenet repository.

  3. Modify pathes in below scripts to fit your own system.

DATA PRCOSSING

  1. Download NTU RGB+D action dataset. Only skeleton data is used here.

  2. Preproess the skeleton data with scripts

TESTING

  1. Download trained models

  2. Extract features of skepxel images with scripts.

  3. Apply FTP on extracted features and split the results according to testing protocols.

  4. Run classification.

TRAINING

  1. Generate skexepl images for with scripts.

  2. Place script to facenet/src.

  3. Config and Run TRAINING script.

References