Foundation Model or Finetune? Evaluation of few-shot semantic segmentation for river pollution
This work addresses the practical choice between foundation models and finetuning for semantic segmentation in environmental monitoring, showing incremental but important results for domain-specific applications.
The paper compared foundation models against finetuned pre-trained supervised models for few-shot semantic segmentation on a new river pollution dataset, finding that finetuned models consistently outperformed foundation models even with scarce data.
Foundation models (FMs) are a popular topic of research in AI. Their ability to generalize to new tasks and datasets without retraining or needing an abundance of data makes them an appealing candidate for applications on specialist datasets. In this work, we compare the performance of FMs to finetuned pre-trained supervised models in the task of semantic segmentation on an entirely new dataset. We see that finetuned models consistently outperform the FMs tested, even in cases were data is scarce. We release the code and dataset for this work on GitHub.