Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PawPrint: Whose Footprints Are These? Identifying Animal Individuals by Their Footprints

Inpyo Song, Hyemin Hwang, Jangwon Lee

TL;DR

This work tackles non-invasive individual animal identification through footprints by introducing PawPrint and PawPrint+, public datasets for dogs and cats across controlled and natural terrains. It benchmarks a range of modern deep networks (CNNs, Vision Transformers) and classical local features (SIFT, ORB) under varied data regimes, highlighting strong performance in controlled conditions and notable declines in heterogeneous environments. The findings suggest that hybrid approaches combining global learned representations with robust local descriptors can improve reliability across real-world scenarios, with domain adaptation as a future avenue. Overall, the datasets and results advance ethical pet management and wildlife conservation by enabling reliable footprint-based identification without invasive sensors.

Abstract

In the United States, as of 2023, pet ownership has reached 66% of households and continues to rise annually. This trend underscores the critical need for effective pet identification and monitoring methods, particularly as nearly 10 million cats and dogs are reported stolen or lost each year. However, traditional methods for finding lost animals like GPS tags or ID photos have limitations-they can be removed, face signal issues, and depend on someone finding and reporting the pet. To address these limitations, we introduce PawPrint and PawPrint+, the first publicly available datasets focused on individual-level footprint identification for dogs and cats. Through comprehensive benchmarking of both modern deep neural networks (e.g., CNN, Transformers) and classical local features, we observe varying advantages and drawbacks depending on substrate complexity and data availability. These insights suggest future directions for combining learned global representations with local descriptors to enhance reliability across diverse, real-world conditions. As this approach provides a non-invasive alternative to traditional ID tags, we anticipate promising applications in ethical pet management and wildlife conservation efforts.

PawPrint: Whose Footprints Are These? Identifying Animal Individuals by Their Footprints

TL;DR

This work tackles non-invasive individual animal identification through footprints by introducing PawPrint and PawPrint+, public datasets for dogs and cats across controlled and natural terrains. It benchmarks a range of modern deep networks (CNNs, Vision Transformers) and classical local features (SIFT, ORB) under varied data regimes, highlighting strong performance in controlled conditions and notable declines in heterogeneous environments. The findings suggest that hybrid approaches combining global learned representations with robust local descriptors can improve reliability across real-world scenarios, with domain adaptation as a future avenue. Overall, the datasets and results advance ethical pet management and wildlife conservation by enabling reliable footprint-based identification without invasive sensors.

Abstract

In the United States, as of 2023, pet ownership has reached 66% of households and continues to rise annually. This trend underscores the critical need for effective pet identification and monitoring methods, particularly as nearly 10 million cats and dogs are reported stolen or lost each year. However, traditional methods for finding lost animals like GPS tags or ID photos have limitations-they can be removed, face signal issues, and depend on someone finding and reporting the pet. To address these limitations, we introduce PawPrint and PawPrint+, the first publicly available datasets focused on individual-level footprint identification for dogs and cats. Through comprehensive benchmarking of both modern deep neural networks (e.g., CNN, Transformers) and classical local features, we observe varying advantages and drawbacks depending on substrate complexity and data availability. These insights suggest future directions for combining learned global representations with local descriptors to enhance reliability across diverse, real-world conditions. As this approach provides a non-invasive alternative to traditional ID tags, we anticipate promising applications in ethical pet management and wildlife conservation efforts.

Paper Structure

This paper contains 22 sections, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. PawPrint Datasets
  4. Dataset Statistics
  5. Train/Test Split.
  6. Footprint Image Collection
  7. Dataset Composition
  8. Animal Identification
  9. Implementation Details
  10. Models and Training.
  11. Hyperparameters.
  12. Quantitative Results
  13. Performance in Controlled Environments
  14. Deep Learning Models.
  15. Local Features.
  16. ...and 7 more sections

Figures (5)

  • Figure 1: Can you correctly match each print to its corresponding individual animals? In this work, we explore a non-invasive, footprint-based identification framework that could benefit both pet owners and wildlife conservation efforts.
  • Figure 2: The proposed PawPrint dataset, with two versions: PawPrint (sand/clay only) andPawPrint+ (additional natural terrains like wood, dirt, snow, rock, asphalt). Our goal is to facilitate individual-level animal identification using their footprints.
  • Figure 3: Sample pictures of dogs and cats from the PawPrint dataset. In total, we collected 2,020 footprint images from 13 dogs and 7 cats across both PawPrint and PawPrint+.
  • Figure 4: PawPrint and PawPrint+ dataset distribution.
  • Figure 5: This figure compares the key point detection of SIFT, the attention maps of ResNet50 Grad-CAM, and ViT. SIFT primarily focuses on key landmarks within the footprint, ResNet50 Grad-CAM highlights the overall shape of the footprint, and ViT captures a more global context of the footprint