Floating Objects Dataset

Citation Author(s):
Liyi
Lei
Submitted by:
Mugle Lei
Last updated:
Sun, 01/27/2019 - 03:22
DOI:
10.21227/69c1-w583
Data Format:
License:
0
0 ratings - Please login to submit your rating.

Abstract 

We have labeled 683 images and 2015 bounding boxes in total for now. We used 80% of the dataset which are 546 images for training and the rest of 20% which are 137 images for evaluation, and all images in the dataset have the same resolution at 1920*1080. Table.I summaries the statistics of the dataset, the category of floating leaves region has the most bounding box labels and the category of floating weeds region has the largest average size of bounding boxes. Since we have only taken pictures at 2 lakes and the number of lotus that grows in the lake is quite small, we only had labeled 58 bounding boxes for lotus. And due to the plants around the lake are not the main object of our subject, we didn’t focus on collecting images that including plants around lakes.

Instructions: 

We used the LabelImg to annotate images, and the annotation files will be saved as XML files in PASCAL VOC format which is the same as ImageNet.