Benchmark:
3D-ZeF20 |
Short name:
Naive
Description:
Utilizes a "Naive" object detection approach in order to detect the fish and their head points. Tracklet creation is performed first in each 2D view and then associated across views. Lastly the tracklets are combined in to full tracks when possible.
Reference:
M. Pedersen, J. Haurum, S. Bengtson, T. Moeslund. 3D-ZeF: A 3D Zebrafish Tracking Benchmark Dataset. In IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020.
Last submitted:
June 16, 2020 (4 years ago)
Published:
June 16, 2020 at 11:44:29 CET
Submissions:
1
Project page / code:
Open source:
Yes
Hardware:
4 GHZ, 1 Core
Runtime:
1.8 Hz
Benchmark performance:
Sequence | MOTA | IDF1 | MOTP | MT | ML | FP | FN | Rcll | Prcn | FAF | ID Sw. | Frag | MTBFm |
3D-ZeF20 | 51.0 | 63.0 | 34.4 | 2 (11.1) | 0 (0.0) | 1,370 | 6,552 | 59.6 | 87.6 | 0.4 | 18 (30.2) | 520 (8.7) | 9.0 |
Detailed performance:
Sequence | MOTA | IDF1 | MOTP | MT | ML | FP | FN | Rcll | Prcn | FAF | ID Sw. | Frag | MTBFm |
ZebraFish-05 | 79.4 | 88.9 | 34.6 | 1 | 0 | 28 | 157 | 82.6 | 96.4 | 0.0 | 0 | 30 | 12.2 |
ZebraFish-06 | 78.4 | 88.2 | 36.6 | 1 | 0 | 45 | 344 | 80.9 | 97.0 | 0.1 | 0 | 44 | 16.2 |
ZebraFish-07 | 40.3 | 54.9 | 32.8 | 0 | 0 | 577 | 2,104 | 53.2 | 80.6 | 0.6 | 6 | 200 | 5.9 |
ZebraFish-08 | 48.0 | 58.6 | 34.6 | 0 | 0 | 720 | 3,947 | 56.1 | 87.5 | 0.8 | 12 | 246 | 9.8 |
Raw data: