Ikea Furniture Assembly Dataset (IkeaFA) is a video dataset. Each video contains a single actor assembling and disassembling an Ikea furniture. The dataset could be helpful for researches on compositional actions and long-term video understanding.

• Size: 101 videos, 1920x1080, 30fps, each 2-4 minutes long.
• Camera control: all videos are captured by a stationary GoPro camera.
• Scene control: furniture assembly task is performed either on a table or on the floor.
• Variance control: 14 actors with different assembly styles and clothing.
• Classes: we define 12 action classes including pick leg, attach leg 1/2/3/4, detach leg 1/2/3/4, spin in/out, flip table. Also a null action is assigned to actions that don't belong to any listed actions above.
• Annotations: We provide two sets of annotations, see the README.txt for details
  1. temporal time stamp for each action
  2. bounding box for the Ikea table

You can download the dataset [here] (9.5GB).

The zip file has following structure:

├── README.txt
├── IkeaClipsDB.mat
├── videos
    ├── 2016-08-11
        ├── GOPR*.MP4
    ├── 2016-08-18
        ├── GOPR*.MP4
    ├── 2016-09-01
        ├── GOPR*.MP4
├── processed-python-data
    ├── ikea_action_data.h5

• [1] Human Action Forecasting by Learning Task Grammars. Tengda Han, Jue Wang, Anoop Cherian, and Stephen Gould (2017). Tech report. [paper]
• [2] Human Pose Forecasting via Deep Markov Models. Sam Toyer, Anoop Cherian, Tengda Han, Stephen Gould (2017). DICTA 2017. [paper]

If you find this dataset useful and use it in published work, please remember to cite the following two papers

@article{han2017,
    title={Human Action Forecasting by Learning Task Grammars},
    author={Han, Tengda and Wang, Jue and Cherian, Anoop and Gould, Stephen},
    journal={arXiv:1709.06391},
    year={2017}}

@inproceedings{toyer2017,
    title={Human Pose Forecasting via Deep {M}arkov Models},
    author={Toyer, Sam and Cherian, Anoop and Han, Tengda and Gould, Stephen},
    booktitle={DICTA},
    year={2017}}

• We highly appreciate the contribution of Andy, Basura, Daniel, Elaine, Guofeng, Han, Harris, Jay, Jue, Kitt, Mike, Rodrigo, Sam and Xinyu for participating in our data collection effort.
• We thank ANU and ACRV for the support of GPUs and furnitures.
• Website last updated at March 2018.