File size: 6,208 Bytes
ebb66db | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 | # UBnormal dataset | [COSKAD](https://github.com/aleflabo/COSKAD) | [Original Repository](https://github.com/lilygeorgescu/UBnormal)
---
Skeletal version proposed in
**[Contracting Skeletal Kinematics for Human-Related Video Anomaly Detection](https://arxiv.org/abs/2301.09489)**</br>
*Alessandro Flaborea\*, Guido D'Amely\*, Stefano D'Arrigo\*, Marco Aurelio Sterpa, Alessio Sampieri, Fabio Galasso*
---
We propose HR-UBnormal as an extension of the original UBnormal dataset [\[1\]]((#references)) with kinematic motion representations and a selected set of anomalies that relate only to human behaviors.
First, AlphaPose [\[2\]](#references) was used to extract the poses, and PoseFlow [\[3\]](#references) was used to track the skeletons throughout each video.
Then, we filtered out the non-human related anomalies. We removed the sub-sequences in which the only anomalous object was not a person (e.g., a car) or the anomaly cannot be detected using only body poses (e.g., fire in the scene).
As a result, we left the validation set unaltered while eliminating the frames 2.32% of the test set.
## Notes regarding the file names' format
The UBnormal dataset annotated with the skeletal representation and its Human-Related version (HR) are released with the following directory structure:
```
UBnormal
|
|__________ hr_bool_masks
| |
| |__________ testing
| | |
| | |__________ test_frame_mask
| | |_______________{scene_id}_{clip_id}.npy
| | |_______________...
| | |_______________{scene_id}_{clip_id}.npy
| |
| |__________ validating
| |
| |__________ test_frame_mask
| |_______________{scene_id}_{clip_id}.npy
| |_______________...
| |_______________{scene_id}_{clip_id}.npy
|
|__________ training
| |
| |__________ trajectories
| |
| |_________{scene_id}_{clip_id}
| |
| |_________00001.csv
| |_________...
| |_________0000{n}.csv
|
|__________ testing
| |
| |__________ trajectories
| | |
| | |_________{scene_id}_{clip_id}
| | |
| | |_________00001.csv
| | |_________...
| | |_________0000{n}.csv
| |
| |__________ test_frame_mask
| |
| |_______________{scene_id}_{clip_id}.npy
| |_______________...
| |_______________{scene_id}_{clip_id}.npy
|
|__________ validating
|
|__________ trajectories
| |
| |_________{scene_id}_{clip_id}
| |
| |_________00001.csv
| |_________...
| |_________0000{n}.csv
|
|__________ test_frame_mask
|
|_______________{scene_id}_{clip_id}.npy
|_______________...
|_______________{scene_id}_{clip_id}.npy
```
In the `hr_bool_masks`, the frames which were anomalous in the original version but where the anomaly didn't involve any human being are toggled to 'normal', i.e., they are toggled from 1 to 0.
Regarding the naming of the files, since our code expects the `scene_id` and the `clip_id` to be integers and some of the file names in the original dataset were overloaded, the following mapping has been adopted:
-
```
scene_id = {c1c2c3}
```
where `c1` is the scene type (`{'abnormal':0, 'normal':1}`) and `c2c3` is the scene number of the corresponding file in the original dataset.
-
```
clip_id = {c1c2c3c4}
```
where `c1c2` is the scenario number (i.e., clip id) of the corresponding file in the original dataset, `c3c4` is the remaining id part dubbed as version. Indeed, in the original dataset some videos have names as in the following example:
- `normal_scene_1_scenario1_1`
- `normal_scene_1_scenario1_10`
- `abnormal_scene_9_scenario_1_fog`
- `abnormal_scene_12_scenario_1_smoke`
In order to keep the information regarding the environment in the clip (e.g., fog, smoke, ...), this mapping has been adopted:
```
{'fog': 51, 'fire': 52, 'smoke': 53}
```
## Citation
If you find this dataset useful in your research, please consider cite:
```
@misc{flaborea2023contracting,
title={Contracting Skeletal Kinematics for Human-Related Video Anomaly Detection},
author={Alessandro Flaborea and Guido D'Amely and Stefano D'Arrigo and Marco Aurelio Sterpa and Alessio Sampieri and Fabio Galasso},
year={2023},
eprint={2301.09489},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
@InProceedings{Acsintoae_CVPR_2022,
author = {Andra Acsintoae and Andrei Florescu and Mariana{-}Iuliana Georgescu and Tudor Mare and Paul Sumedrea and Radu Tudor Ionescu and Fahad Shahbaz Khan and Mubarak Shah},
title = {UBnormal: New Benchmark for Supervised Open-Set Video Anomaly Detection},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2022},
}
```
## References
[1] A. Acsintoae, A. Florescu, M.-I. Georgescu, T. Mare, P. Sumedrea, R. T. Ionescu, F. S. Khan, M. Shah, Ubnormal: New benchmark for supervised open-set video anomaly detection, in: Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022, pp.20143–20153.
[2] H.-S. Fang, S. Xie, Y.-W. Tai, C. Lu, Rmpe: Regional multi-person pose estimation, in: ICCV, 2017, pp. 2334–2343.
[3] Y. Xiu, J. Li, H. Wang, Y. Fang, C. Lu, Pose Flow: Efficient online pose tracking, in: BMVC, 2018.
|