The Zebrafish challenge (3D-ZeF20). This benchmark proposes a new challenging addition to the Multi-Object Tracking benchmarks, by extending it to 3D tracking of zebrafish swimming in a laboratory environment. We recorded 8 challenging video sequences (4 training, 4 test) with a stereo vision setup and annotated with bounding boxes and head points. Tracking and evaluation are done in 3D coordinates. All sequences have been annotated with high accuracy, following a well-defined protocol. The sequences ZebraFish-03 and ZebraFish-04 were used as a validation set in the original paper.

Training Set

Sample Name FPS Resolution Length Tracks BoxesDensityDescriptionSourceRef.
ZebraFish-04602704x1520910 (00:15)5910010.05 zebrafish (VAL5)link[1]
ZebraFish-03602704x15201800 (00:30)272004.02 zebrafish (VAL2)link[1]
ZebraFish-02602704x1520900 (00:15)5900010.05 zebrafish (TRN5)link[1]
ZebraFish-01602704x15207188 (02:00)2287524.02 zebrafish (TRN2)link[1]
Total 10798 frm.
(180 s.)
14 54052 5.0

Test Set

Sample Name FPS Resolution Length Tracks BoxesDensityDescriptionSourceRef.
ZebraFish-08602704x1520900 (00:15)101800020.010 zebrafishlink[1]
ZebraFish-07602704x1520900 (00:15)5900010.05 zebrafishlink[1]
ZebraFish-06602704x1520900 (00:15)236004.02 zebrafishlink[1]
ZebraFish-05602704x1520900 (00:15)118002.01 zebrafishlink[1]
Total 3600 frm.
(60 s.)
18 32400 9.0


Get all data (14.0 GB)
Get detections and labels only (0.8 MB)
Get development kit (0.5 MB)


[1] Pedersen, M., Haurum, J.B., Bengtson, S.H. & Moeslund, T.B. 3D-ZeF: A 3D Zebrafish Tracking Benchmark Dataset. arXiv:2006.08466[cs], 2020., (arXiv: 2006.08466).