Dataset overview

The BovineAAEyes80 is a dataset comprising 80 RGB images of 11 different bovines belonging to Aberdeen-Angus breed. The dataset is specially suited for detection/segmentation of iris/pupil regions of the bovine's eye, fundamental problems in automatic biometric recognition of animals.

For each RGB image, there are three accompanying ground-truth label pictures, indicating the pixels that belong to three possible regions:

• Outer border pixels (pixels inside the closed area delimited by iris' outer border, i.e. iris + pupil pixels),
• Inner border pixels (pixels inside the closed area delimited by iris' inner border, i.e. pupil pixels),
• Iris ROI pixels (outer border pixels minus inner border pixels, i.e. iris pixels).

Each image shows the bovine's head pictured from the left or right side. The photos have been taken at night using a regular visible light camera, employing the camera's built-in flash. The capture distance from the camera to the animal varies between 100 cm and 200 cm.

Image resolution is 2100x1575 pixels for all images. Each picture is labeled as <animal number>_<image number>, with animal number ranging from 1 to 11, and image number starting at 1 for each different bovine.

Files organization

There is a separate subfolder for each of the four types of images:

• RGB images are at data/rgb/,
• Outer Border labels are at data/labels_outer_border/,
• Inner border labels are at data/labels_inner_border/,
• ROI labels are at data/labels_roi/.

For example: RGB image 5_1.png can be found at subfolder data/rgb/5_1.png, whilst the corresponding ROI labels image is located at data/labels_roi/5_1.png.

You can download the dataset (~500MB) or view the project on Github.

Results of our segmentation method

A table of the results achieved by our segmentation method, for every image, can be seen in this link. The evaluation metrics are Precision, Recall, F1 (Dice) and IoU (Jaccard index).

Citations

If you find this dataset useful for your research, please cite our work:

@article{larregui2019image, title={An image processing pipeline to segment iris for unconstrained cow identification system}, author={Larregui, Juan I and Cazzato, Dario and Castro, Silvia M}, journal={Open Computer Science}, volume={9}, number={1}, pages={145--159}, year={2019}, publisher={De Gruyter} }