Data available for "Identification of herbarium specimen sheet components from high-resolution images using deep learning": Annotations for selected MELU specimen sheet digital images

Data Available for the paper: "Identification of herbarium specimen sheet components from high-resolution images using deep learning", by Karen M Thompson, Robert Turnbull, Emily Fitzgerald, Joanne L Birch

These are specific annotations of selected specimen sheet digital images from the MELU collection (Melbourne University Herbarium). MELU collection images are available: https://online.herbarium.unimelb.edu.au/ 

These annotations for use in a YOLO object detection model.

The format of this file is a .ZIP containing a .TXT for each image annotated. 

Each .TXT file will have a row for each annotated element.

Eg. "4 0.064133 0.414363 0.072186 0.309392"

(i) first element is an integer identifying the object type:

 0 small database label

 1 handwritten data
2 stamp
3 annotation label
4 scale
5 swing tag
6 full database label
7 database label
8 swatch
9 institutional label
10 number

(ii) then the following four elements are the corner coordinates for the bounding box

Other information available to support this paper:

(1) annotations for benchmark dataset (noting these are specific to the MELU trained model) 

(2) MELU-trained sheet-component object detection model weights (for application in YOLOv5)