PlantTrack: Task-Driven Plant Keypoint Tracking with Zero-Shot Sim2Real Transfer

TL;DR

PlantTrack leverages foundation models and synthetic training data to enable zero-shot plant feature tracking in real environments, addressing challenges posed by unstructured and deformable plant scenes.

Contribution

We introduce PlantTrack, a method combining DINOv2 and TAPIR for zero-shot plant keypoint tracking using minimal synthetic data for training.

Findings

Achieves effective zero-shot transfer with only 20 synthetic images.

Successfully tracks plant features in real-world environments.

Utilizes foundation models for robust feature detection and tracking.

Abstract

Tracking plant features is crucial for various agricultural tasks like phenotyping, pruning, or harvesting, but the unstructured, cluttered, and deformable nature of plant environments makes it a challenging task. In this context, the recent advancements in foundational models show promise in addressing this challenge. In our work, we propose PlantTrack where we utilize DINOv2 which provides high-dimensional features, and train a keypoint heatmap predictor network to identify the locations of semantic features such as fruits and leaves which are then used as prompts for point tracking across video frames using TAPIR. We show that with as few as 20 synthetic images for training the keypoint predictor, we achieve zero-shot Sim2Real transfer, enabling effective tracking of plant features in real environments.