PlantCLEF 2025

2025

A CVPR challenge in computer vision: name every plant species in a photo of a quadrat, having only ever been trained on pictures of one plant at a time.

  • ML
Stack
Python · PyTorch · DINOv3

Problem

The training images are single-label close-ups; the test images are quadrats with several species tangled together. A model trained on the former does not transfer cleanly to the latter, and closing that gap is the whole challenge.

What I built

Multi-label classification models built on top of DINOv3, targeting that domain shift between the training and test distributions.