Sample frames for this project are available under NDA. Write to bilal@wortel.ai and we will share the relevant before and after pairs.
Challenges
- Detecting animals within dense natural environments.
- Identifying small or distant animals inside an image.
- Handling animals partly hidden behind vegetation or other objects.
- Separating animals from background elements such as rocks, trees and shadows.
- Varied animal poses and orientations complicated bounding box placement.
- Lighting conditions and image quality affected how visible each animal was.
- Keeping bounding boxes tight and accurate around animals demanded careful attention to detail.
Related work
Poultry Chick Counting
Bounding boxes on individual chicks for automated counting, including crowded groups where birds overlap heavily.
Engineering Drawing Segmentation
Rooms, service runs and annotation blocks marked out on CAD floor plans so drawings can be parsed automatically.
Ground Cover Quadrat Segmentation
2,275 cover-crop plot images annotated in Labelbox: plot corners boxed, then every pixel classed as groundcover, weed, residue, crop or soil.