Justin KayI am a PhD student at MIT, advised by Sara Beery and supported by fellowships from MIT EECS and NSF. My research focuses on making computer vision and machine learning systems more deployable and informative for science and decision-making, particularly for environmental and climate applications. Before coming to MIT, I co-founded Ai.Fish, a computer vision company focused on applications in sustainable fisheries management and ocean conservation. From 2021–2023 I was also a researcher in the Computational Vision Lab at Caltech, advised by Pietro Perona. I hold a BS in EECS from UC Berkeley. Email / GitHub / Google Scholar / LinkedIn |
|
Recent News |
| 2025-04-01: | Co-organizing the 2025 AI upskilling workshop for former US/state environmental scientists with Dan Morris, Sara Beery, and Eric Orenstein. |
| 2025-03-15: | ALDI accepted to TMLR with a Featured Certification! Check out the paper, codebase, and dataset. |
| 2025-02-04: | Teaching Computer Vision for Planetary Health with Sara Beery. |
| 2024-10-28: | Invited talk at Valle University in conjunction with CBD COP16 in Cali. |
| 2024-09-25: | Invited talk at EU DG-MARE Fisheries Science Seminar in Brussels. |
| 2023-12-16: | Spotlight Talk at the Computational Sustainability Workshop at NeurIPS. Recording available here. |
| 2023-10-13: | Invited talk at the Korean Ministry of Fisheries AI and Electronic Monitoring Expert Seminar in Seoul. |
| 2023-10-05: | Organized the Exploring AI for Fisheries in Depth workshop at SAFET in Bali. A recording of my talk, "Fair Evaluation of AI systems," is available here. |
| 2023-08-01: | Ai.Fish was awarded a NOAA SBIR Phase II grant for our project "Cloud-based Automated Electronic Monitoring for Fisheries of the Future"! I will serve as Co-Investigator. Read more here. |
Selected Publications |
|
Align and Distill: Unifying and Improving Domain Adaptive Object DetectionJustin Kay, Timm Haucke, Suzanne Stathatos, Siqi Deng, Erik Young, Pietro Perona, Sara Beery, and Grant Van Horn Transactions on Machine Learning Research, 2025 ☆ Featured Certification (Spotlight), TMLR ☆ Best poster honorable mention, NECV 2023 website / arxiv / code / A unified framework for domain adaptive object detection that addresses systemic benchmarking pitfalls and achieves state-of-the-art performance across diverse benchmarks. |
|
Unsupervised Domain Adaptation in the Real World: A Case Study in Sonar VideoJustin Kay, Suzanne Stathatos, Siqi Deng, Erik Young, Pietro Perona, Sara Beery, and Grant Van Horn Computational Sustainability Workshop at NeurIPS, 2023 ☆ Best paper, pitfalls track website / presentation / Highlighting the promises and pitfalls of unsupervised domain adaptation through a real-world case study in fisheries monitoring. |
|
The Caltech Fish Counting Dataset: A Benchmark for Multiple-Object Tracking and CountingJustin Kay, Peter Kulits, Suzanne Stathatos, Siqi Deng, Erik Young, Sara Beery, Grant Van Horn, and Pietro Perona European Conference on Computer Vision (ECCV), 2022 website / arxiv / poster / presentation / A large-scale dataset for detecting, tracking, and counting fish in sonar videos, an important conservation application and a rich data source for advancing low signal-to-noise computer vision applications and tackling domain generalization in tracking and counting. |
|
Fine-Grained Counting with Crowd-Sourced SupervisionJustin Kay, Catherine M. Foley, and Tom Hart Computer Vision for Animal Behavior Tracking and Modeling Workshop at CVPR, 2022 website / arxiv / poster / A dataset and method for fine-grained (i.e. multiclass) seal counting using noisy volunteer dot annotations collected from the Zooniverse citizen science platform. |
|
The Fishnet Open Images Database: A Dataset for Fish Detection and Fine-Grained Categorization in FisheriesJustin Kay and Matt Merrifield 8th Workshop on Fine-Grained Visual Categorization at CVPR, 2021 website / arxiv / poster / A large and diverse image dataset sourced from fisheries electronic monitoring. Evaluation of existing detection and classification algorithms and quantification of key challenges. |
|
Design and source code from here. |