USGS Izembek result previews

This page hosts results for sample patches for the Izembek bird detector v1.0.0. All results shown here are either validation data (patches selected from images excluded from training, but from flights that were included in training) or test data (patches selected from flights that were excluded from training), but all are from the same years that were used for model training. The test data represents a realistic data distribution (with lots of blanks); the validation data is biased toward images that contain animals (though most individual patches are still empty). Hence the ~83% blank rate in the validation data, compared to the ~98% blank rate in the test data. Previews are based on a single confidence threshold (0.625); the goal is just to communicate the gist of what the model is doing, rather than to facilitate deep exploration of model output.

• Previews for validation data (an assortment of images held out from training)
• Previews for test data (2017_replicate_2017-10-03)

All images are sampled from:

Weiser, E. L., Flint, P. L., Marks, D. K., Shults, B. S., Wilson, H. M., Thompson, S. J. and Fischer, J. B., 2022, Aerial photo imagery from fall waterfowl surveys, Izembek Lagoon, Alaska, 2017-2019: U.S. Geological Survey data release, https://doi.org/10.5066/P9UHP1LE.