You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
some labels are non-informative due to tiling, like this example where natural seep is identified in the upper left corner but appears as a small dot where there isn't much signal.
The text was updated successfully, but these errors were encountered:
Now that we have the stats defined in #44 , we can look at the distribution of area by each category and come up with an area threshold we think will address most of the artifacts introduced due to tiling. We could also potentially get fancier with this filter and filter out small area annotations that occur at the scene edges. @lillythomas assigning you for now but we can discuss when you're back
I think we should apply this filter after parsing the coco dataset by the icevision trainer and fastai trainers. so the function to do this should operate on a numpy array, with another function to handle dealing with the icevision record or fastai2 sample.
some labels are non-informative due to tiling, like this example where natural seep is identified in the upper left corner but appears as a small dot where there isn't much signal.
The text was updated successfully, but these errors were encountered: