Open devoworm opened 7 years ago
We could do something like the following: We create a testing dataset by manually marking cell boundaries in some images(At least 10-20 different images). The marked cell boundaries will establish the ground truth.
Then, we form some kind of loss function(we could start with Euclidean Distance) to determine differences between the result given by our algorithms and the ground truth.
Sid will get back with Bradly on issue. Feels comfortable with implementation.
Possible approaches: