Skip to content

Latest commit

 

History

History
38 lines (22 loc) · 1.23 KB

README.md

File metadata and controls

38 lines (22 loc) · 1.23 KB

VIBE Data parsing scripts

This folder contains script which can be used to parse the VIBE skeleton data of skeletics-152 and skeleton-mimetics

There is one main script:

  1. parse_vibe_data.py (Parses VIBE data into VaCNN and GCN format)
  • VaCNN data is stored in .h5 format
  • GCN data is stored in .npy format

GCN data structure:

GCN data is stored is most standard format used by most of the GCN based models.

The data is stored in N x C x T x V x M , where

  • N = Number of samples
  • C = Number of channels (C = 3 for 3d data)
  • T = Number of Frames (T = 300 by default)
  • V = Number of Joints (V = 25 by default)
  • M = Number of Persons (M = 2 by default)

How to run script

python3 parse_vibe_data.py INPUT_PATH OUTPUT_PATH

Where

  • INPUT_PATH = path to the directory containing VIBE .json files
  • OUTPUT_PATH = path where parsed data needs to be saved.

After the script is finished a vacnn_data.h5 file will be created with VaCNN data and labels.

Also two more files will be created, gcn_train_data.npy and gcn_train_label.npy which will contain data and label respectively.

(NOTE: To parse different splits of dataset give the input path accordingly)