Date of Graduation


Document Type


Degree Name

Bachelor of Science in Computer Science


Computer Science and Computer Engineering


Le, Thi Hoang Ngan

Committee Member/Reader

Gauch, Susan

Committee Member/Second Reader

Nakarmi, Ukash


This honors thesis dives into the realm of deep learning-based pose estimation research and investigates the potential of DeepLabCut (Lauer, et al., 2021) in accurately and efficiently estimating the pose of poultry. With accurate pose estimation being a crucial aspect in understanding the behavior and movement of animals, this thesis aims to contribute to the development of more effective methods for pose estimation, especially for poultry.

To comprehensively evaluate the performance of DeepLabCut, two different types of chickens were tested in this thesis: a model toy chicken and actual live chickens. Videos were recorded for both types, and key points were manually labeled for selected frames. This labeling process served as the foundation for the creation of a DeepLabCut model, which was trained on the dataset and then evaluated for its performance on a separate validation dataset.

The result of this thesis showcases the good capabilities of DeepLabCut in accurately and efficiently estimating the pose of poultry when provided with sufficient data. The trained models were able to accurately predict the pose of the chickens in the videos when the models had sufficient training data on the poses. However, due to insufficient data, at certain poses with insufficient training data, the model that was created for the live chickens was overall not great performance-wise.

To enhance the model's performance, a selective approach was employed to increase the size of the training dataset by focusing on troublesome body parts and poses that had insufficient training data. In the end, the model's overall performance demonstrated improvement, although the increase was modest.


Poultry, DeepLabCut, Computer, Pose-Estimation