xuxie1031/unsupervisedgloveaction

Unsupervised Learning of Hierarchical Models for Hand-Object Interactions (ICRA 2018)

Unsupervised Learning of Hierarchical Models for Hand-Object Interactions (ICRA 2018)

Source Code of Clustering pipeline / ACA optimization / Inference labeling

Usage

$ git clone https://github.com/xiaozhuchacha/AtomicAction.git

Clustering

• use 'main.m' as the entry in 'Clustering' folder to start clustering pipeline
• sample hand data is placed in 'Clustering/hand_data'

ACA

• use 'ACA.cpp' as the entry in 'ACA' folder to start clustering optimization
• to compile: g++ ACA.cpp -o ACA
• execute: ./ACA [NC] [data_name], eg. ./ACA 9 sample_hand_data

Inference

• use 'AnnealGibbs.py' as the entry in 'Inference' folder to start Gibbs annealing
• parser is placed in 'Inference/induced_grammar' to calculate prior
• gaussians are placed in 'Inference/Gaussians' to calculate likelihood
• use 'main.m' as the entry in 'Inference/Gaussians' to compute gauusian parameters w.r.t ground truth labeled data
• labeled motion sequence as input can be found either in 'Clustering/hcluster' as hierarchical clustering result or 'ACA/ACAbin' as ACA optimization result
• execute: python AnnealGibbs.py [--nlabel] [--data-name], eg. python AnnealGibbs.py --nlabel 6 --data-name sample_hand_data

Contact

• [email protected]

Paper Link

• see gloveaction2018icra.pdf