A deep learning project in cooperation with the NOAA Marine Mammal Lab to detect & classify arctic seals in aerial imagery to understand how they’re adapting to a changing world.
MIT License
33
stars
14
forks
source link
Fixed RGB bounding boxes for most of the seal hotspots. Training on … #31
Fixed 99% of labels for rgb bounding boxes for seal hotspots.. The new labels are already giving me much better localization and classification results. The new status column is "new" if it's a new label that didn't exist before and "removed" if it's a bad label or an image with terrible quality. "new" labels are based on the original hotspot id in the image but with an additional .1, .2, .n... to the id and the other metadata was copied from the original hotspot. The IR center points have not been updated at all this is only for rgb boxes. I will continue cleaning this file up to make a very good training set but for now I figure this could be useful to anyone working on this project. If you have any questions feel free to contact me, all this was done in my Bears-n-Seals repo.
Fixed 99% of labels for rgb bounding boxes for seal hotspots.. The new labels are already giving me much better localization and classification results. The new status column is "new" if it's a new label that didn't exist before and "removed" if it's a bad label or an image with terrible quality. "new" labels are based on the original hotspot id in the image but with an additional .1, .2, .n... to the id and the other metadata was copied from the original hotspot. The IR center points have not been updated at all this is only for rgb boxes. I will continue cleaning this file up to make a very good training set but for now I figure this could be useful to anyone working on this project. If you have any questions feel free to contact me, all this was done in my Bears-n-Seals repo.