Skip to content

Hawaii Model

The Hawaii model has been created using almost 51 hours of soundscape data containing 59,583 bounding box labels representing 27 different species.


Training Results

The following figure illustrates various data that has been recorded during training. The metrics have been computed after each epoch with the evaluation split of the dataset.

Training Results

Model Checkpoint

The weights that produced the highest metrics/mAP50-95 on the validation split were selected for the final model.


Species Distribution Across Splits

The following table shows the amount of annotations in total and for each species as described in Dataset Splits.

Species Train Val Test Total 70/15/15 Quality
Total 63,261 13,388 13,562 90,211 70.1/14.8/15.0
akepa1 962 199 209 1,370 70.2/14.5/15.3
apapan 7,265 1,544 1,482 10,291 70.6/15.0/14.4
barpet 598 117 130 845 70.8/13.8/15.4
calqua 89 19 20 128 69.5/14.8/15.6
elepai 219 49 40 308 71.1/15.9/13.0
ercfra 5,269 1,162 1,282 7,713 68.3/15.1/16.6
hawama 10,616 2,157 2,251 15,024 70.7/14.4/15.0
hawcre 897 230 194 1,321 67.9/17.4/14.7
hawhaw 254 51 54 359 70.8/14.2/15.0
hawpet1 790 188 193 1,171 67.5/16.1/16.5
houfin 5,011 1,110 997 7,118 70.4/15.6/14.0
iiwi 8,695 1,847 1,800 12,342 70.5/15.0/14.6
jabwar 422 96 61 579 72.9/16.6/10.5
melthr 1,209 226 283 1,718 70.4/13.2/16.5
norcar 231 57 51 339 68.1/16.8/15.0
omao 2,812 580 643 4,035 69.7/14.4/15.9
reblei 6,093 1,247 1,333 8,673 70.3/14.4/15.4
skylar 7,457 1,584 1,603 10,644 70.1/14.9/15.1
warwhe1 3,889 849 866 5,604 69.4/15.1/15.5
wiltur 77 19 18 114 67.5/16.7/15.8
yefcan 242 57 52 351 68.9/16.2/14.8
blkfra 6 0 0 6 Train-only
chukar 86 0 0 86 Train-only
comwax 15 0 0 15 Train-only
hawgoo 3 0 0 3 Train-only
kalphe 17 0 0 17 Train-only
palila 37 0 0 37 Train-only