Skip to content

Latest commit

 

History

History
45 lines (32 loc) · 1.8 KB

README.md

File metadata and controls

45 lines (32 loc) · 1.8 KB

Data Generation and Augmentation

This folder contains tools to generate diverse scene-aware or object-aware human motion data using existing datasets.

Locomotion in 3D-FRONT Scene

Data Sources

HumanML3D Dataset

  1. Follow instructions in HumanML3D
  2. Copy the dataset to our repository: 
    cp -r ../HumanML3D/HumanML3D ./dataset/HumanML3D
  3. Set $HUMANML_3D_ROOT to the HumanML3D dataset folder

AMASS Dataset

  1. Download from AMASS
  2. Set $AMASS_DATA to the dataset folder

3D-FRONT Data

  1. Download from 3D-FRONT
  2. Set $threeDFront_root to the dataset folder.
  3. Get the bird-view floor plan and object mask for each scene, you can refer to this scripts.

Fitting Scripts

  • Fitting script: data_generation/locomotion/align_motion_amass.py

Human-Object Interaction

Based on Summon, we predict contact areas for each motion frame and fit objects of corresponding categories.

Data Sources

  1. SAMP dataset: Download from SAMP and set $DATA_ROOT/SAMP
  2. 3D-FUTURE dataset:
    • Download from 3D-FUTURE
    • Set $DATA_ROOT/3D-FUTURE-model
    • We use raw_model.obj for each subject

Fitting Scripts

  1. Fit objects to predicted contact areas: data_generation/interaction/summon/fit_best_obj.py
  2. Calculate transform matrix and merge into .pkl: data_generation/interaction/summon/sort_out_result.py
  3. Visualization: data_generation/interaction/summon/vis_fitting_results.py