Hi @threehundred, I have noticed discrepencies in the bounding boxes from FDFML/labelled/Manifests and FDFML/labelled/Speciesboxes for a large number of frames that come from group E videos.
From your Github page, I presumed the only difference was that some of the labels in FDFML/labelled/Speciesboxes also have species labells attached. I am wondering which group of labels should be trusted more for the purposes of building a Machine Learning algorithm? From quick inspection, it seems the labels in FDFML/labelled/Speciesboxes have been somewhat cleaned, removing some false positives?
As an example of the discrepencies, here are the bounding boxes from FDFML/labelled/Manifests for frame E000596_L.MP4.16074.png:
Whilst here are the bounding boxes from FDFML/labelled/speciesboxes for frame E000596_L.MP4.16074.png:
Hi @threehundred, I have noticed discrepencies in the bounding boxes from FDFML/labelled/Manifests and FDFML/labelled/Speciesboxes for a large number of frames that come from group E videos.
From your Github page, I presumed the only difference was that some of the labels in FDFML/labelled/Speciesboxes also have species labells attached. I am wondering which group of labels should be trusted more for the purposes of building a Machine Learning algorithm? From quick inspection, it seems the labels in FDFML/labelled/Speciesboxes have been somewhat cleaned, removing some false positives?
As an example of the discrepencies, here are the bounding boxes from FDFML/labelled/Manifests for frame E000596_L.MP4.16074.png:
Whilst here are the bounding boxes from FDFML/labelled/speciesboxes for frame E000596_L.MP4.16074.png: