NID Dataset

This folder contains the NID dataset prepared for the evaluation of insect detection and localization algorithms. If you use this dataset, please cite the corresponding paper:

D. Korsch, P. Bodesheim, G. Brehm, and J. Denzler. Automated Visual Monitoring of Nocturnal Insects with Light-based Camera Traps. CVPR Workshop on Fine-grained Visual Classification (CVPR-WS), 2022.

For more details visit the website of the dataset: https://inf-cv.uni-jena.de/nid_dataset/

Directory Information

• images/ This directory contains subdirectories named after the corresponding moth species and each subdirectory contains the sample images of that species

• images.txt The filenames of the images relative to the images directory Each line contains a separate image.

• multi_boxes.json Bounding box annotations for the insects in json format. For each image from images.txt, there is a list of objects and each object is specified by box coordinates (x0,y0) and (x1,y1) with ranges between 0 and 1 relative to the image size.

• tr_ID.txt Proposed train-test-split for each image in images.txt. Each line in tr_ID.txt corresponds to the image in the same line in images.txt. A value of 0 means that this image is used for testing and a value of 1 means this image is used for training.