Segmenting On Other Forms Of Data Instead
Hi,
I came across this vegetation segmentation model and I’m interested in applying it to pasture/grass imagery for a downstream task (biomass estimation).
I wanted to ask whether this model generalizes well to fine-grained grass or pasture vegetation, or if it would require domain-specific fine-tuning to maintain high accuracy in such settings.
Have you (or others) evaluated its performance specifically on grassland or agricultural pasture images? Any guidance on expected limitations or best practices for adaptation would be very helpful.
Thanks for sharing the model and for any insights.
Hello,
Thanks for your query.
I am not aware of any specific grass segmentation projects utilizing the model, as it was created more for "in leaf" shrub and tree vegetation.
That being said, the model may segment grasses well with only a handful of training data samples (RGB, multi-spectral) using LoRA or other adaptive techniques.
Rural areas with large swaths of healthy grasses, intermixed with trees will probably segment well with minimal training data verses a full fine-tune. Seasonal dried grasses could also be a target for the model.
Let me know if I might assist with any follow up questions.
MR