search

vanvalenlab / deepcell-tracking

Track cells and build cell lineages
Other
29 stars 6 forks source link

Identify how divisions in t +/- 1 frame are handled in the metrics package #101

Closed msschwartz21 closed 1 year ago

msschwartz21 commented 2 years ago

Changes in segmentation predictions can lead to a shift in the frame in which one cell is split into two new cells. Even if the division is accurately identified it may occur in a slightly different frame. Our metrics should not penalize this behavior and should identify it as a correct division as long as it falls within a few frames of the ground truth division.

msschwartz21 commented 2 years ago

Currently we use the frame information when we are mapping between nodes in the GT and the predictions: https://github.com/vanvalenlab/deepcell-tracking/blob/ea01e7569942addf6c87d89f2476800917ef06d9/deepcell_tracking/metrics.py#L52-L55

I think that this means that we will not identify divisions that occur in different frames. I suspect that we may need an entirely different matching strategy to identify divisions that are assigned to a slightly different frame.