Spectrogram segmentation for underwater noise classification

We use transfer learning on yolov7 to detect bounding boxes around potential dolphin whistles.

Results on REPMUS datasets

Data Type	Training Labels	Precision	Recall	mAP@0.5
REP23	430	0.59	0.50	0.31
REP23+aug	1300	0.66	0.69	0.55
REP23+21+aug	5000	0.79	0.69	0.77