Key Features

As an end-to-end deep learning platform which focus on bird detection in RGB images, WaterfowlDetector has advantages for detecting small-bird objects in high resolution aerial images. These tricks can help you get better detection performance without changing the deep learning model networks themselves.

Datasets with high diversity

Across years of data collection, we collected images of birds and decoys from different conservation areas in Missouri and generated ten datasets (see table below. These images were taken under different light conditions across different vegetation communities, seasons, waterfowl densities, and waterfowl species. This diversity in our imagery allowed us to develop robust models that are able to detect and classify waterfowl among many complex backgrounds.

Prebuilt models

We developed six pretrained detection and X pretrained classification models based on our datasets. These models are trained by different datasets using different images to provide better performance for specific scenarios (see table below). We also provide some general models which are trained by part or all the datasets for combined or complex datasets which cover a range of scenarios.