Animal Vision AI (AV-AI)
Last updated
Last updated
Subject images submitted through WILD dApp are processed using AV-AI, a state-of-the-art biometric scanning and image processing system that combines advanced facial recognition, pattern recognition, and posture analysis. AV-AI utilizes a dual deep-learning approach:
FaceNet512 is an advanced facial recognition model based on the Inception-ResNet-v1 architecture, which combines concepts from both Inception networks and Residual networks. Originally designed for human facial recognition, FaceNet512 has been adapted and fine-tuned for animal facial feature extraction. It generates a 512-dimensional embedding vector that captures unique identifying characteristics of animal faces. This high-dimensional representation allows for precise differentiation between individual animals, even within the same species or breed.
CLIP is a powerful vision transformer model that excels in general visual feature extraction and potential breed classification.
This combination allows for comprehensive animal identification, encompassing facial features, body patterns, and structural elements. Pose estimation models are integrated to analyze body structure and posture, providing additional identifying characteristics.
The system processes each image to extract a rich, multi-faceted feature vector. This vector, along with associated metadata, is stored in a high-performance vector database, enabling efficient similarity searches and rapid retrieval for animal identification and matching.
Once the AV-AI has successfully processed the images, a unique digital identifier (DID) is generated for the subject animal based on the subject’s biometric patterns and stored for future reference.
Users complete a series of prompts designed to capture attributes relating to the subject animal that are not available to the biometric scanning output, such as age, sex, home address, etc. Additionally, users complete data entries relating to preferences such as pet food, medical issues, treatment routines, vaccination records, etc.
AV-AI is designed to perform a subject recognition sequence to determine whether the subject is already stored in the database. For example, Bob finds a dog on the roadside. The dog doesn’t have a collar with owner-specific identification. Bob takes the dog to the animal shelter. Animal shelter staff scan the dog’s microchip, which has been placed but is no longer functional. As an alternative, the shelter scans the animal using WILD Platform’s AV-AI. The images are processed, and a DID has already been assigned to the dog. The owner Sally’s contact information is available on the network, and the shelter calls her to come and retrieve her pet.
