Need to attempt to locate growth points. This would allow estimation of pitcher lengths.
Possibly useful heuristics:
- centroid of convex hull
- dark regions ringed by more or less linear features (assumes pitchers aren't bent/occluded)
- skeletonization, then find root node
It would be much easier to train a neural network for this than for pitcher segmentation, since labeling here involves only one or two point locations per image, rather than painstaking pitcher tracing. Maybe worth a shot too?
Need to attempt to locate growth points. This would allow estimation of pitcher lengths.
Possibly useful heuristics:
It would be much easier to train a neural network for this than for pitcher segmentation, since labeling here involves only one or two point locations per image, rather than painstaking pitcher tracing. Maybe worth a shot too?