Large_Fragment Label Policy
Within the dataset there are a number of areas that contain prolonged image-errors caused by damage to the EM. Usually they manifest as consecutive misaligned or torn slices that disrupt your ability to proofread. We've found that it is common for neurons to have backbones pass through these bad areas, making their continuation difficult and time consuming, if not altogether improbable. We have been roughly annotating these bad areas, creating a map that, among other things, allows us to speed up the labeling process. Here is a neuroglancer link with the bounding box tab. See the definition of backbone.
The policy is, if the neuron you are labeling has a major backbone that enters one of these bounding boxes and ends within it, label it as a "large_fragment" (and proofread the rest of the neuron). If a major backbone ends just outside a bounding box, inspect it. These branches are likely broken due to the outer reaches of the image error, which are more often than not, brief disruptions rather than prolonged disruptions. If you determine that the image error prevents continuation, label it as "large_fragment" (and proofread the rest of the neuron). If a backbone enters and leaves the bounding box intact, proofread it as best you can while still sticking to normal time policy, but do not apply the large_fragment label. INon-major backbones ending within the bounding boxes do not earn the large_fragment label.
If you come across an additional bad area that isn't represented in the map, kindly share the coordinates and a link with Ben Silverman or Kyle Willie on the Flywire slack. Currently we are updating how we annotate these areas. Soon we’ll have meshes that more accurately adhere to the actual parameters of the errors.
Summary
Gallery
