Back to Works

Instance Segmentation

Instance Segmentation of Person/Phone/Credit Card and select randomize object.

Instance Segmentation

The Challenge

The challenge is a part of public competition by CSIRO

Farmers often walk into a paddock and ask one question: “Is there enough grass here for the herd?” It sounds simple, but the answer is anything but. Pasture biomass - the amount of feed available - shapes when animals can graze, when fields need a break, and how to keep pastures productive season after season.

Current methods make this assessment more challenging than it could be. The old-school “clip and weigh” method is accurate but slow and impossible at scale. Plate meters and capacitance meters can provide quicker readings, but are unreliable in variable conditions. Remote sensing enables broad-scale monitoring, but it still requires manual validation and can’t separate biomass by species.

Our Approach

With the help of Meta's pretrained SSL model DINOV3, we fine-tuned the model on the given dataset.

Tech Stack

Python
PyTorch
Kafka

Key Features

  • ConvNets
  • Transfer Learning
  • Model Ensembling