Our Publications
Reprints are available upon request
Zero-shot Shark Tracking and Biometrics from Aerial Imagery
Chinmay Lalgudi, Mark Leone, Jaden Clark, Sergio Madrigal-Mora, Mario Espinoza
Methods in Ecology and Evolution, 2025
We demonstrate that pre-trained models (FLAIR, a custom architecture built with SAM2 and CLIP) can effectively track arbitrary shark species in various environments and be used to extract key biometrics. We benchmark against fine-tuned object detection and segmentation models and find that pre-trained models outperform task-specific models on our nurse shark dataset.
Deep Learning for Automated Shark Detection and Biometrics Without Keypoints
Jaden Clark*, Chinmay Lalgudi*, Mark Leone*, Sergio Madrigal-Mora, Mario Espinoza
ECCV Computer Vision for Ecology, 2024
We show a simple method for computing shark biometrics from aerial drone imagery using custom fine-tuned object detectors to prompt Segment Anything Model (SAM). By applying heuristics for centerline estimation and photogrammetry, we estimate shark length from SAM masks.
