Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

DogFLW: Dog Facial Landmarks in the Wild Dataset

George Martvel, Greta Abele, Annika Bremhorst, Chiara Canori, Nareed Farhat, Giulia Pedretti, Ilan Shimshoni, Anna Zamansky

TL;DR

DogFLW introduces a 46-point canine facial landmark dataset—DogFLW—comprising 3,274 in-the-wild images across 120 breeds, with bounding boxes and visibility flags for each landmark. The landmark scheme is grounded in dog facial musculature and DogFACS, and annotations were produced via a human-in-the-loop approach to drastically reduce labeling time. Benchmarks using the Ensemble Landmark Detector (ELD) with YOLOv8 and EfficientNetV2S show a full-training $NME_{iod}$ of 6.52, with ear regions and certain breeds posing substantial challenges. The work demonstrates the feasibility and value of robust canine facial analysis for emotion and welfare research, while identifying data diversity—particularly ear types and long-fur breeds—as essential directions for future improvements. Overall, DogFLW provides a foundational resource to advance canine affective computing and welfare monitoring through AI-driven landmark detection.

Abstract

Affective computing for animals is a rapidly expanding research area that is going deeper than automated movement tracking to address animal internal states, like pain and emotions. Facial expressions can serve to communicate information about these states in mammals. However, unlike human-related studies, there is a significant shortage of datasets that would enable the automated analysis of animal facial expressions. Inspired by the recently introduced Cat Facial Landmarks in the Wild dataset, presenting cat faces annotated with 48 facial anatomy-based landmarks, in this paper, we develop an analogous dataset containing 3,274 annotated images of dogs. Our dataset is based on a scheme of 46 facial anatomy-based landmarks. The DogFLW dataset is available from the corresponding author upon a reasonable request.

DogFLW: Dog Facial Landmarks in the Wild Dataset

TL;DR

DogFLW introduces a 46-point canine facial landmark dataset—DogFLW—comprising 3,274 in-the-wild images across 120 breeds, with bounding boxes and visibility flags for each landmark. The landmark scheme is grounded in dog facial musculature and DogFACS, and annotations were produced via a human-in-the-loop approach to drastically reduce labeling time. Benchmarks using the Ensemble Landmark Detector (ELD) with YOLOv8 and EfficientNetV2S show a full-training of 6.52, with ear regions and certain breeds posing substantial challenges. The work demonstrates the feasibility and value of robust canine facial analysis for emotion and welfare research, while identifying data diversity—particularly ear types and long-fur breeds—as essential directions for future improvements. Overall, DogFLW provides a foundational resource to advance canine affective computing and welfare monitoring through AI-driven landmark detection.

Abstract

Affective computing for animals is a rapidly expanding research area that is going deeper than automated movement tracking to address animal internal states, like pain and emotions. Facial expressions can serve to communicate information about these states in mammals. However, unlike human-related studies, there is a significant shortage of datasets that would enable the automated analysis of animal facial expressions. Inspired by the recently introduced Cat Facial Landmarks in the Wild dataset, presenting cat faces annotated with 48 facial anatomy-based landmarks, in this paper, we develop an analogous dataset containing 3,274 annotated images of dogs. Our dataset is based on a scheme of 46 facial anatomy-based landmarks. The DogFLW dataset is available from the corresponding author upon a reasonable request.
Paper Structure (8 sections, 1 equation, 5 figures, 3 tables)

This paper contains 8 sections, 1 equation, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Landmark Scheme
  3. Dataset Properties
  4. Benchmarks
  5. Dataset Size
  6. Ear Types
  7. Breeds
  8. Conclusion

Figures (5)

  • Figure 1: Annotated Dog's Face. Image of a dog with a face bounding box and 46 facial landmarks.
  • Figure 2: Examples of annotated images from the DogFLW.
  • Figure 3: Impact of the training subset size on the normalised mean error. The size of the subset is measured as a fraction of the total size of the training data.
  • Figure 4: Landmark detection examples on pre-cropped images from the test set. Red --- ground truth, blue --- predictions.
  • Figure 5: The distribution of images in the training set for each breed of dogs and the detection errors for the same breeds in the test set.