---
license: mit
task_categories:
- video-classification
language:
- en
tags:
- action-recognition
- temporal-actions
- video-recognition
- video-representations
size_categories:
- 10K<n<100K
configs:
- config_name: default
  data_files:
  - split: ssv2_train
    path: "chiral_groups-ssv2-train.csv"
  - split: ssv2_validation
    path: "chiral_groups-ssv2-validation.csv"
  - split: epic_train
    path: "chiral_groups-epic-train.csv"
  - split: epic_validation
    path: "chiral_groups-epic-validation.csv"
  - split: charades_train
    path: "chiral_groups-charades-train.csv"
  - split: charades_validation
    path: "chiral_groups-charades-validation.csv"
---


# [Chirality in Action: Time-Aware Video Representation Learning by Latent Straightening](https://arxiv.org/abs/2509.08502)

ArXiv: https://arxiv.org/abs/2509.08502

We provide all the splits for the three datasets uses in chiral action recognition proposed in the paper.

The description of each split is as follows:

| Split | Description |
|-------|-------------|
| `chiral_groups-combined.csv` | Combined chiral groups for all datasets |
| `chiral_groups-ssv2-train.csv` | Chiral groups for SSv2 train set |
| `chiral_groups-charades-train.csv` | Chiral groups for Charades train set |
| `chiral_groups-epic-train.csv` | Chiral groups for Epic train set |
| `chiral_groups-ssv2-validation.csv` | Chiral groups for SSv2 validation set |
| `chiral_groups-charades-validation.csv` | Chiral groups for Charades validation set |
| `chiral_groups-epic-validation.csv` | Chiral groups for Epic validation set |
| `cia-ssv2-train.csv` | Chiral groups for SSV2 train set |
| `cia-ssv2-validation.csv` | Chiral groups for SSV2 validation set |
| `cia-charades-train.csv` | Chiral groups for Charades train set |
| `cia-charades-validation.csv` | Chiral groups for Charades validation set |
| `cia-epic-train.csv` | Chiral groups for Epic train set |
| `cia-epic-validation.csv` | Chiral groups for Epic validation set |


The splits with prefix `chiral_groups` contain chiral groups: 
(verb positive, verb negative, noun) triplets.
The splits with prefix `cia` contain the train and validation 
splits for each dataset with only the subset of the dataset 
relevant for chiral action recognition.

If you have any questions, please contact us via email.

#### Citation

If you use this dataset in your research, please cite the following paper:
```
@article{bagad2025chirality,
  title={Chirality in Action: Time-Aware Video Representation Learning by Latent Straightening},
  author={Bagad, Piyush and Zisserman, Andrew},
  journal={arXiv preprint arXiv:2509.08502},
  year={2025}
}
```