Figure 6
From: FathomNet: A global image database for enabling artificial intelligence in the ocean
Results from the NOAA Benthic use case. (a) Validation confusion matrix from the RetinaNet object detector trained on supercategory data from Monterey Bay. The row of background false negatives is mostly due to incomplete original annotation data. (b) Confusion matrix from the model applied to NOAA data from the Musician Seamount. Note that the vast majority of the targets were sea fans. (c) Model output from a single frame of NOAA video data. (d) Ground truth annotations for the same frame. Note the density, variability in coral morphology, and overlapping individuals.
