Optimal Dataset Scale & Optimal Image Scale metric
For anyone who has worked on image segmentation, can you please explain to me what these 2 metrics mean ? I am still quite unsure of how they are computed. I cannot find a paper that specifically mentions what they are.
From Contours to Regions: An Empirical Evaluation https://vision.ics.uci.edu/papers/Arb...
Currently this is how I think the ODS, OIS scores are generated (not specific to that paper). Can someone please confirm ?
1. Run edge detector on image, getting an edge map
2. Given the ground truth, get the precision-recall scores with recall from 0 to max
3. Get the F scores for each precision-recall score
4. Optimal dataset scale is the threshold at which the F score is maxed across the dataset
5. Optimal image scale is the threshold at which the F score is maxed across each image