LAC: Latent Action Composition for Skeleton-based Action Segmentation

Project Page | Paper

This is the official PyTorch implementation of the ICCV 2023 paper "LAC: Latent Action Composition for Skeleton-based Action Segmentation"

Installation

Clone this repo:

git clone https://github.com/walker1126/Latent_Action_Composition.git
cd Latent_Action_Composition

Install dependencies:

pip install -r requirements.txt

Action Generator (for Skeleton Action Composition)

Generator training (by motion retargeting):

python train_generator.py -n view -g 0

Inference (action composition):

python predict.py -n view --model_path ./model/pretrained_view.pth -v1 ./examples/walk.json -v2 ./examples/drink.json -h1 720 -w1 720 -h2 720 -w2 720 -o ./outputs/com-demo --max_length 60

Skeleton Contrastive Learning & Action Segmentation

Comming soon!

@InProceedings{Yang_2023_ICCV,
    author    = {Yang, Di and Wang, Yaohui and Dantcheva, Antitza and Kong, Quan and Garattoni, Lorenzo and Francesca, Gianpiero and Bremond, Francois},
    title     = {LAC - Latent Action Composition for Skeleton-based Action Segmentation},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {13679-13690}
}