WiserZhou/ProcedurePlanning

Masked Temporal Interpolation Diffusion (MTID) Dataset for Procedure Planning

This repository contains the datasets used in the paper Masked Temporal Interpolation Diffusion for Procedure Planning in Instructional Videos.

The MTID (Masked Temporal Interpolation Diffusion) model addresses the challenge of procedure planning in instructional videos. It aims to generate coherent and task-aligned action sequences from start and end visual observations by leveraging a latent space temporal interpolation module to augment visual supervision with richer mid-state details. This dataset facilitates research and development in this area by providing necessary data for training and evaluating such models.

The code for the MTID model is available at: https://github.com/WiserZhou/MTID

Data Preparation

This dataset includes data for three widely used benchmark datasets: CrossTask, COIN, and NIV.

To download datasets and features, navigate to the respective dataset directory and run the download script as shown in the original repository:

cd ./dataset/{dataset_name}
bash download.sh

Replace {dataset_name} with crosstask, coin, or NIV.

Alternatively, you can find the datasets within this Hugging Face repository itself.

Citation

If you find this dataset or the associated paper useful in your research, please cite:

@inproceedings{
  zhou2025masked,
  title={Masked Temporal Interpolation Diffusion for Procedure Planning in Instructional Videos},
  author={Yufan Zhou and Zhaobo Qi and Lingshuai Lin and Junqi Jing and Tingting Chai and Beichen Zhang and Shuhui Wang and Weigang Zhang},
  booktitle={ICLR},
  year={2025},
}