Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

PetFace: A Large-Scale Dataset and Benchmark for Animal Identification

Risa Shinoda, Kaede Shiohara

TL;DR

PetFace tackles the challenge of automatic animal face identification by introducing a large-scale, richly annotated dataset with 257,484 individuals across 13 animal families and 319 breeds, totaling 1,012,934 images. It establishes two benchmarks—re-identification for seen individuals and verification for unseen individuals—and demonstrates that ArcFace-based methods, especially with joint-training on PetFace, achieve strong performance and generalization across unseen families and breeds. The dataset includes fine-grained attributes (sex, breed, color/pattern) and employs rigorous alignment and filtering to maintain quality. Overall, PetFace advances animal face identification by enabling cross-family evaluation, improving generalization, and providing a resource that can catalyze non-invasive identification methods for real-world applications.

Abstract

Automated animal face identification plays a crucial role in the monitoring of behaviors, conducting of surveys, and finding of lost animals. Despite the advancements in human face identification, the lack of datasets and benchmarks in the animal domain has impeded progress. In this paper, we introduce the PetFace dataset, a comprehensive resource for animal face identification encompassing 257,484 unique individuals across 13 animal families and 319 breed categories, including both experimental and pet animals. This large-scale collection of individuals facilitates the investigation of unseen animal face verification, an area that has not been sufficiently explored in existing datasets due to the limited number of individuals. Moreover, PetFace also has fine-grained annotations such as sex, breed, color, and pattern. We provide multiple benchmarks including re-identification for seen individuals and verification for unseen individuals. The models trained on our dataset outperform those trained on prior datasets, even for detailed breed variations and unseen animal families. Our result also indicates that there is some room to improve the performance of integrated identification on multiple animal families. We hope the PetFace dataset will facilitate animal face identification and encourage the development of non-invasive animal automatic identification methods.

PetFace: A Large-Scale Dataset and Benchmark for Animal Identification

TL;DR

PetFace tackles the challenge of automatic animal face identification by introducing a large-scale, richly annotated dataset with 257,484 individuals across 13 animal families and 319 breeds, totaling 1,012,934 images. It establishes two benchmarks—re-identification for seen individuals and verification for unseen individuals—and demonstrates that ArcFace-based methods, especially with joint-training on PetFace, achieve strong performance and generalization across unseen families and breeds. The dataset includes fine-grained attributes (sex, breed, color/pattern) and employs rigorous alignment and filtering to maintain quality. Overall, PetFace advances animal face identification by enabling cross-family evaluation, improving generalization, and providing a resource that can catalyze non-invasive identification methods for real-world applications.

Abstract

Automated animal face identification plays a crucial role in the monitoring of behaviors, conducting of surveys, and finding of lost animals. Despite the advancements in human face identification, the lack of datasets and benchmarks in the animal domain has impeded progress. In this paper, we introduce the PetFace dataset, a comprehensive resource for animal face identification encompassing 257,484 unique individuals across 13 animal families and 319 breed categories, including both experimental and pet animals. This large-scale collection of individuals facilitates the investigation of unseen animal face verification, an area that has not been sufficiently explored in existing datasets due to the limited number of individuals. Moreover, PetFace also has fine-grained annotations such as sex, breed, color, and pattern. We provide multiple benchmarks including re-identification for seen individuals and verification for unseen individuals. The models trained on our dataset outperform those trained on prior datasets, even for detailed breed variations and unseen animal families. Our result also indicates that there is some room to improve the performance of integrated identification on multiple animal families. We hope the PetFace dataset will facilitate animal face identification and encourage the development of non-invasive animal automatic identification methods.
Paper Structure (19 sections, 12 figures, 5 tables)

This paper contains 19 sections, 12 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. PetFace Dataset
  4. Dataset Statistics
  5. Data Sourcing
  6. Face Alignment and Filtering
  7. Fine-Grained Annotations
  8. Experimental Methodology
  9. Evaluation Protocols
  10. Data Split
  11. Models
  12. Experimental Result
  13. Benchmark on Animal Face Re-identification
  14. Benchmark on Animal Face Verification
  15. Comparison with Previous Datasets
  16. ...and 4 more sections

Figures (12)

  • Figure 1: Example images of our PetFace. We introduce a large-scale animal face re-identification dataset PetFace that include 257,484 unique individuals across 13 families and 319 breeds. From the left, the images represent Cat, Chimpanzee, Chinchilla, Degus, Dog, Ferret, Guinea pig, Hamster, Hedgehog, Parakeet, Java sparrow, Pig, and Rabbit. The four images enclosed within the white square's grid lines represent the same identity.
  • Figure 2: Dataset distribution. Each figure represents (a) the number of individuals in each animal family. (b) the sex distribution percentage by animal family. (c) examples of breed annotations. (d) examples of color annotations.
  • Figure 3: Data filtering process. We adopt a two-stage data filtering approach. Initially, images containing multiple faces are automatically removed. Subsequently, any images that do not depict animal faces or fail to meet alignment criteria are manually eliminated.
  • Figure 4: Top-1
  • Figure 5: Top-3
  • ...and 7 more figures