The official codebase of FineAction dataset. We will update the data and code of our FineAction.

[Arxiv version] [TIP version]

ToDo

• Update the VideoMAE feature on FineAction.
• Update the code of reproduce codabases(BMN, ActionFormer) on FineAction.

📰 News

• [2022-10-10]: Our FineAction has been accepted by IEEE Transactions on Image Processing.

• [2022-04-30]: We have published competitions for FineAction on DeeperAction Workshop@ECCV2022 Track 1.

• [2021-05-31]: We have published competitions for FineAction on DeeperAction Workshop@ICCV2021 Track 1.

⚡️ Data Download

• Raw videos: You should obey the terms and conditions and download them from our new competition pages.

• Annotations: BaiduYun (Access Code: 64vp) or Onedrive

• I3d feature: BaiduYun (Access Code: vbf0) or Onedrive

• I3d_100 feature: BaiduYun (Access Code: v2rp) or Onedrive

• VideoMAE feature: [BaiduYun] or [Onedrive]

🚀 Research

We will update the SOTA methods on our FineAction.

✨ Temporal Action Localization

Year	Venue	Model (or Authors)	@0.5	@0.75	@0.95	AVG	code
2019	ICCV	BMN (Reprod.)	14.44	8.92	3.12	9.25	code
2020	AAAI	DBG (Reprod.)	10.65	6.43	2.50	6.75	code
2020	CVPR	G-TAD (Reprod.)	13.74	8.83	3.06	9.06	code

✨ Weakly Supervised Temporal Action Localization

Year	Venue	Model (or Authors)	@0.5	@0.75	@0.95	AVG	code
2018	ECCV	W-TALC (Reprod.)	6.18	3.15	0.83	3.45	code
2021	AAAI	Lee et al. (Reprod.)	6.65	3.23	0.95	3.64	code
2021	CVPR	ASL (Reprod.)	6.79	2.68	0.81	3.30	code
2021	ICCV	D2-Net (Reprod.)	6.75	3.02	0.82	3.35	code
2022	ECCV	HAAN	7.05	3.95	1.14	4.10	code

☎️ Contact

Any question about our FineAction, you can email me. ([email protected])

✏️ Citation

If you think this project is helpful, please feel free to leave a star⭐️ and cite our paper:

@article{liu2022fineaction,
  title={Fineaction: A fine-grained video dataset for temporal action localization},
  author={Liu, Yi and Wang, Limin and Wang, Yali and Ma, Xiao and Qiao, Yu},
  journal={IEEE Transactions on Image Processing},
  year={2022},
  publisher={IEEE}
}