Dice coefficient is used as evaluaion metric to measure the segmentation accuracy of bird ROI detection.
Dice coefficient = (area of intersection between original ROI and detected ROI)/sum of area of original+detected ROIs. The value of dice coef varies from 0 to 1.
Each image contains a single bounding box label.
boundingbox_gt.txt contain groundtruth bounding box labels and boundingbox_pred.txt contain the detected boundingbox labels, with each line corresponding to one image in the format:
<image_id> <x> <y> <width> <height>
where <image_id> corresponds to the ID of each image, and <x>, <y>, <width>, and <height> are all measured in pixels
"Blue" - groundtruth bounding box
"green" - detected bounding box
"red" - ROI intersection of groundtruth and detected bounding box
/dice_coef_eval_codes - codes and files for dice coefficient evalaution
/bounding_box_samples - examples for groundtruth ROIs
/dice_coef_eval_example - different cases of dice coefficient estimation
/multiple_bird_detection- examples of multiple bird detection cases