Animal Re-Identification on the PetFace dataset

Motivation

Animal re-identification (re-ID) is a crucial component in various applications, from wildlife monitoring to pet recovery. While significant progress has been made in human face recognition, animal re-ID presents unique challenges due to greater intra-species variations and the complexity of animal facial features. The recent release of the PetFace dataset, containing over 257,000 unique individuals across 13 animal families, provides an opportunity to advance the field of animal re-ID and develop more robust and generalizable solutions.

Thesis

Working on this thesis will start with conducting a brief literature review of re-identification methods. The next steps are reproducing the results of one animal re-identification method from the PetFace paper.

Based on that, several research questions could be explored:

• Can we perform transfer learning between common classes in PetFaces (e.g. cat and dog) to uncommon classes (e.g. chimp)?
• Does pretraining on PetFaces improve the performance of re-ID models on different datasets?
• Can we improve the state-of-the-art performance on animal re-ID using different model architectures or training data choices?

Contact

To apply please email Felix Benjamin Müller stating your interest in this project and detailing your relevant skills. Please note that this project is better suited for a student who already has a basic familiarity with deep learning and pytorch.