Datasets¶
Here we provide information and download links for datasets used for training and evaluating DREEM across a range of tracking scenarios. Datasets are organized into two categories: Animals (behavioral tracking of whole organisms) and Microscopy (tracking of cells and subcellular structures). All datasets, including metadata, are available on Google Drive. See below for links to individual datasets.
Hint
Need a quick testing clip? Animals: A short, simple video with four flies. Microscopy: A short clip with slow moving cell nuclei.
Summary¶
Animals¶
| Dataset | Subject | Animals | Videos | Frames |
|---|---|---|---|---|
mice_btc |
Mice | 2 | 25 | ~1.05M |
mice_hc |
Mice | 2 | 36 | ~31K |
slap2m |
Mice | 1–4 | 17 | ~294K |
flies13 |
Fruit Flies | 2–8 | 58 | ~94K |
zebrafish10 |
Zebrafish | 10 | 18 | 27K |
Microscopy¶
| Dataset | Subject | Objects per Frame | Videos | Frames |
|---|---|---|---|---|
lysosomes |
Lysosomes | 1–8 | 10 | ~3.9K |
dynamicnuclearnet |
Cell Nuclei | 3–249 | 130 | ~6.7K |
motchallenge_dic |
Cells (DIC) | 0–117 | 37 | ~80K |
mouse_c2c12 |
Mouse C2C12 Cells | 2–95 | 19 | ~20K |
Animals¶
mice_btc¶
| Name | mice_btc |
|---|---|
| Description | Pairs of mice (Mus musculus) continuously monitored in a behavioral tracking chamber. Long-duration recordings captured simultaneously from multiple camera angles at high frame rates. |
| Videos | 25 (19 train / 3 val / 3 test) |
| Image size | 768 x 1024 x 1 |
| Num Animals | 2 |
| Frames | ~1.05M |
| Download | Google Drive |
| Credit | Talmolab, Salk Institute for Biological Studies |
mice_hc¶
| Name | mice_hc |
|---|---|
| Description | Pairs of mice (Mus musculus) in a home cage setting, imaged from above. Short clips extracted from longer social interaction recordings. Animals can be low contrast against the bedding background. |
| Videos | 36 (32 train / 3 val / 1 test) |
| Image size | 1024 x 1280 x 1 |
| Num Animals | 2 |
| Frames | ~31K |
| Download | Google Drive |
| Credit | Pereira, T. D. et al. SLEAP: A deep learning system for multi-animal pose tracking. Nat. Methods 19, 486–495 (2022). |
slap2m¶
| Name | slap2m |
|---|---|
| Description | Mice (Mus musculus) tracked using the SLAP2 two-photon imaging rig. Variable group sizes from single animals up to groups of 4, with long-duration continuous recordings. |
| Videos | 17 (11 train / 2 val / 4 test) |
| Image size | 1024 x 1280 x 1 |
| Num Animals | 1–4 |
| Frames | ~294K |
| Download | Google Drive |
| Credit | Faulkner Lab, Princeton University |
flies13¶
| Name | flies13 |
|---|---|
| Description | Groups of 2, 4, or 8 freely interacting fruit flies (Drosophila melanogaster) in circular arenas. Contains three sub-conditions (two-flies, four-flies, eight-flies) representing different group sizes. Annotated with a 13-point skeleton with identity. |
| Videos | 58 (46 train / 7 val / 5 test) |
| Image size | 1024 x 1024 x 1 |
| Num Animals | 2–8 |
| Frames | ~94K |
| Download | Google Drive |
| Credit | Pereira, T. D. et al. SLEAP: A deep learning system for multi-animal pose tracking. Nat. Methods 19, 486–495 (2022). |
zebrafish10¶
| Name | zebrafish10 |
|---|---|
| Description | Groups of 10 zebrafish (Danio rerio) freely swimming, imaged at very high spatial resolution. All clips contain exactly 10 tracked individuals throughout. |
| Videos | 18 (6 train / 6 val / 6 test) |
| Image size | 3712 x 3712 x 1 |
| Num Animals | 10 |
| Frames | 27K |
| Download | Google Drive |
| Credit | Romero-Ferrero, F., Bergomi, M. G., Hinz, R. C., Heras, F. J. H. & de Polavieja, G. G. idtracker.ai: tracking all individuals in small or large collectives of unmarked animals. Nat. Methods 16, 179–182 (2019) |
Microscopy¶
lysosomes¶
| Name | lysosomes |
|---|---|
| Description | Lysosomal organelles in live cells imaged with Airyscan confocal microscopy. Small fields of view with a variable number of organelles per video. Organelles exhibit rapid, non-linear motion. |
| Videos | 10 (3 train / 1 val / 6 test) |
| Image size | 250 x 250 x 1 |
| Num Objects | 1–8 per frame |
| Frames | ~3.9K |
| Download | Google Drive |
| Credit | ManorLab, University of California, San Diego |
dynamicnuclearnet¶
| Name | dynamicnuclearnet |
|---|---|
| Description | Fluorescently labeled cell nuclei from the DynamicNuclearNet tracking benchmark. Spans a wide range of cell densities and mitotic activity across multiple experimental conditions. |
| Videos | 130 (91 train / 27 val / 12 test) |
| Image size | 592 x 608 x 1 |
| Num Objects | 3–249 per frame |
| Frames | ~6.7K |
| Download | Google Drive |
| Credit | Schwartz, M. S. et al. Caliban: Accurate cell tracking and lineage construction in live-cell imaging experiments with deep learning. bioRxiv 803205 (2019). |
motchallenge¶
| Name | motchallenge |
|---|---|
| Description | Cell tracking sequences from the MOT Challenge benchmark imaged with Differential Interference Contrast (DIC) microscopy. Spans multiple mammalian cell lines with varying morphology and density, including dividing and migrating cells. |
| Videos | 37 (27 train / 3 val / 7 test) |
| Image size | 320 x 400 x 1 |
| Num Objects | 0–117 per frame |
| Frames | ~80K |
| Download | Google Drive |
| Credit | Anjum, S. & Gurari, D. Ctmc: Cell tracking with mitosis detection dataset challenge 982–983 (2020). |
phase_contrast¶
| Name | phase_contrast |
|---|---|
| Description | Mouse C2C12 myoblast cells imaged with phase contrast microscopy over long recording periods. High-resolution images with many cells per frame undergoing mitosis and migration. |
| Videos | 19 (14 train / 1 val / 4 test) |
| Image size | 1040 x 1392 x 1 |
| Num Objects | 2–95 per frame |
| Frames | ~20K |
| Download | Google Drive |
| Credit | Ker, D. F. E. et al. Phase contrast time-lapse microscopy datasets with automated and manual cell tracking annotations. Sci. Data 5, 180237 (2018) |