Justin Kay

FGVC's not dead!

The 13th Workshop on Fine-Grained Visual Categorization has been accepted to CVPR 2026, in Denver, Colorado!

CALL FOR PAPERS: sites.google.com/view/fgvc13/

From Ecology to Medical Imagining, join us as we tackle the long tail and the limits of visual discrimination! #CVPR2026 #AI

How do we do more with biodiversity data we've already collected?

I gave a TED Talk on scientific discovery in ecological databases at a joint TED Countdown and Bezos Earth Fund event for #NYClimateWeek this year, and it's now live!

@inaturalist.bsky.social #AIforConservation

CleverBirds: A Multiple-Choice Benchmark for Fine-grained Human Knowledge Tracing Mastering fine-grained visual recognition, essential in many expert domains, can require that specialists undergo years of dedicated training. Modeling the progression of such expertize in humans rema...

From medicine to geo-guessing, humans can get incredibly good at solving visual recognition tasks.
But how is this skill learned, and can we model its progression?
We present CleverBirds, accepted #NeurIPS2025, a large-scale benchmark for visual knowledge tracing.
📄 arxiv.org/abs/2511.08512
1/5

Wrapping up @iccv.bsky.social with an in-person CODA demo session! Try it yourself here if you missed us: t.co/yUWp8UCg47

Are you coming to @iccv.bsky.social??

Do you like BIRDS?!?! The #CVBirdWalk is happening!!!

Oct. 21, 7am, Ala Moana Park. Kick off your conference right!!

#ICCV2025 #AIforConservation

Join us tomorrow for the Joint Workshop on Marine Vision at #ICCV2025 (@iccv.bsky.social), from 9-6 in Room 318B!

Check out the full program here: vap.aau.dk/marinevision/

Interested in doing a PhD in machine learning at the University of Edinburgh starting Sept 2026?

My group works on topics in vision, machine learning, and AI for climate.

For more information and details on how to get in touch, please check out my website:
homepages.inf.ed.ac.uk/omacaod

This new work by @justin-kay.bsky.social provides ML users with an efficient mechanism to wade through the sea of available open-source models and identify which model will work best for their needs!

Check it out below, and learn more at our @iccv.bsky.social poster (highlight!) on 10/21!!

Consensus-Driven Active Model Selection The widespread availability of off-the-shelf machine learning models poses a challenge: which model, of the many available candidates, should be chosen for a given data analysis task? This question of...

CODA is a Highlight at @iccv.bsky.social next week!
Poster Session 1 Tuesday 10/21 11:45am, and Demo Session 6 Thursday 10/23 at 2:30pm.

Paper: www.arxiv.org/abs/2507.23771
Demo: huggingface.co/spaces/justi...
Code: github.com/justinkay/coda
@csail.mit.edu News:
bit.ly/48s6yS2

8/8

Thanks to MIT CSAIL @csail.mit.edu for covering CODA and how it supports ML practitioners in environmental conservation: t.co/SNJnnGfKvn

Try it yourself! We built a fun demo for finding the best wildlife classification model with @hf.co and @gradio-hf.bsky.social: huggingface.co/spaces/justi...

7/

CODA is exceptionally label-efficient. On a benchmark suite of 26 different datasets, we show that CODA identifies the optimal or near-optimal model with fewer than 25 labeled examples over 50% of the time, and with fewer than 100 labeled examples over 80% of the time. 6/

CODA constructs a probabilistic model of which model is best at any labeling budget. To do this, we estimate confusion matrices for each candidate that we can 1) integrate over at any time to estimate which is best, and 2) update with new labels via Bayesian inference. 5/

To do this efficiently, CODA leverages the *wisdom of the crowd* (of AI models), using the ensemble predictions from candidate models as a prior over the true labels of your unlabeled data. This helps identify high-value data points where top-performing models disagree. 4/

Typically, answering the model selection question requires collecting a large test dataset to determine which candidate model is the best for you. CODA instead makes the process *active* – interactive, iterative, and guided by the models themselves. 3/

Using AI for data analysis has typically meant training your own model. Large public model zoos like @hf.co Models are changing this paradigm, but pose a new challenge: which model, of the millions available, should you use to analyze your data? 2/

There are now millions of publicly-available AI models – which one is right for you?

We introduce CODA ( #ICCV2025 Highlight! ), a method for *active model selection.* CODA selects the best model for your data with any labeling budget – often as few as 25 labeled examples. 1/

@iccv.bsky.social

Join us at the Joint Workshop for Marine Vision at #ICCV25

We have full-paper and nectar tracks.
Accepted papers are included in the official ICCV workshop proceedings.
Submission deadline: June 30

More at: vap.aau.dk/marinevision/

@iccv.bsky.social
@jbhaurum.bsky.social
@justin-kay.bsky.social

The call for applications has just been released for #CV4Ecology2026!! This three-week intensive program trains ecologists and conservation practitioners to develop their own AI tools for their own data.

When: Jan 12-30, 2026
Where: SCBI @smconservation.bsky.social

There will be a workshop on underwater computer vision at @iccv.bsky.social with papers included in the proceedings (approx. deadline ultimo June) - more information coming soon! @justin-kay.bsky.social @jbhaurum.bsky.social
@aicentre.dk

Machine Learning Evaluation - A Syllabus What we read this semester, and what worked, and what didn't.

A syllabus for a course on machine learning evaluation.

Have you ever wondered what makes two models different?
We all know the ViT-Large performs better than the Resnet-50, but what visual concepts drive this difference? Our new ICLR 2025 paper addresses this question! nkondapa.github.io/rsvc-page/

BirdCLEF25: Audio-based species identification focused on birds, amphibians, mammals, and insects in Colombia.
👉 www.kaggle.com/competitions...
@cvprconference.bsky.social @kaggle.com
#FGVC #CVPR #CVPR2025 #LifeCLEF
[1/4]

Project webpage: aldi-daod.github.io
Paper: arxiv.org/abs/2403.12029
Codebase (consider giving us a star!): github.com/justinkay/aldi
Dataset: github.com/visipedia/ca...

Team: @timm.haucke.xyz, Suzanne Stathatos, Siqi Deng, Erik Young, Pietro Perona, @sarameghanbeery.bsky.social, and Grant Van Horn

Contributions:
- SOTA results across architectures (Faster R-CNN, DETR, YOLO), backbones (ResNet, ConvNeXt, ViT), datasets
- Unified benchmarking & implementation framework, making it easy to develop and test new adaptation methods
- A new real-world adaptation dataset sourced from fisheries sonar

Adapt object detectors to new data *without labels* with Align and Distill (ALDI), our domain adaptation framework published last week in Transactions on Machine Learning Research (with a Featured Certification [Spotlight]!)

Overview of the competitions hosted at the FGVC12: Fine-grained visual categorization workshop at CVPR 2025

Looking for a challenge? Along with FGVC12 at #CVPR2025, multiple fine-grained computer vision competitions have been launched. These well-curated datasets allow you to leverage your AI skills to boost applications with impact!
Let's go: sites.google.com/view/fgvc12/...
@cvprconference.bsky.social

The Computer Vision for Ecology program is organizing a 2-week remote workshop in May to teach AI methods to former US public sector scientists - Please consider sharing with your networks! We are also looking for volunteer instructors/mentors. Details here: cv4ecology.caltech.edu/ai-upskillin...

FGVC12 Workshop at CVPR 2025 - Call for papers.

FGVC12 Workshop accepted to CVPR 2025, Nashville!
CALL FOR PAPERS: sites.google.com/view/fgvc12/...
We discuss domains where expert knowledge is typically required and investigate artificial systems that can efficiently distinguish a large number of very similar visual concepts.
#CVPR #CVPR2025 #AI

🙋‍♂️