Carolina Bay Object Detector Images and Labels

Contained here (in CarolinaBayTrainingData.zip) are the images (jpegs) and labels (csv) that were used to train a bounding box object detector for Carolina Bays detailed in the following paper:

Lundine, M., Trembanis, A., Using Convolutional Neural Networks for Detection and Morphometric Analysis of Carolina Bays from Publicly Available Digital Elevation Models, Remote Sensing, 2021, Volume 13(18), 3770, https://doi.org/10.3390/rs13183770.

Each image is a single channel jpeg. The channel corresponds to gridded LiDAR elevation values remapped to 256 values. Each image was normalized individually for maximum contrast.

The labels csv contains the coordinates for each bounding box annotation of Carolina Bays in each available image. The columns are: filename, width, height, label, xmin, ymin, xmax, ymax, label_value. Each row is an annotation of a Carolina Bay. Filename corresponds to the image, width is the width in pixels of that image, height is the height in pixels of that image. xmin, ymin, xmax, ymax are the bounding box coordinates for the annotation. Label is 'bay' for each annotation, with a label_value of 1.

Feel free to experiment with this dataset, add to it, and improve upon the results.

Also contained here (in CarolinaBayDetections.zip) are the detection results (as unaggregated polygons, as aggregated polygons, as smooth polygons, and as points).