Table 3. Ground truth content, by dataset.
Table lists ground truth properties for each dataset used in this work to evaluate the model. None of the datasets listed here were used for training. ‘Vol’ refers to volume and ‘indiv’ refers to individuals. Ground truth ‘matches pair−1’ indicates the average number of ground truth matches for random pairs of test and template, which is a property of the ground truth dataset, and does not depend on the model tested.
| Held-out semi-synthetic testset | Ca2+ imaging | NeuroPAL | NeuroPAL | Ca2+ imaging | |
|---|---|---|---|---|---|
| Figure | Figure 2 | Figure 3 | Figures 4 and 5 | Figure 4F | Figure 3—figure supplement 1 |
| Type | - | moving | immobile | immobile | moving |
| Correspondence | across indiv | within indiv | across indiv | across indiv | within indiv |
| Ground Truth | simulator | human | human | human | - |
| Ground truth matches pair−1 | 85.7 | 64.4 | 50.1 | 50.5 | - |
| Ground truth labels vol−1 | 102.1 | 69.2 | 57.5 | 64.3 | - |
| Segmented neurons vol−1 | 114.1 | 118.4 | 133.6 | 118.8 | 131.1 |
| Total Vols | 2000 | 1514 | 11 | 9 | 1400 |
| Individuals | 2000 | 1 | 11 | 9 | 1 |
| Vols indiv−1 | 1 | 1514 | 1 | 1 | 1400 |
| Vols s−1 | - | 6 | - | - | 6 |
| Strain | - | AML32 | AML320 (via OH15262) | OH15495 | AML310 |
| Reference | this work | Nguyen et al., 2017 | this work | Chaudhary et al., 2021 | Hallinen et al., 2021 |