InsectSAM: Insect Segmentation and Monitoring

Overview

InsectSAM is a fine-tuned version of Meta AI's segment-anything model, optimized for insect segmentation and monitoring in the Netherlands. Designed for use with the DIOPSIS camera systems, algorithms and datasets, it enhances the accuracy of insect biodiversity segmentation from complex backgrounds.

Purpose

Trained to segment insects against diverse backgrounds, InsectSAM adapts to changing environments, ensuring its long-term utility for the DIOPSIS datasets.

Model Architecture

Built on the segment-anything architecture, InsectSAM is fine-tuned on an insect-specific dataset and integrated with GroundingDINO for improved detail recognition.

Quick Start

Prerequisites

• Python
• Hugging Face Transformers
• PyTorch

Usage

Install

!pip install --upgrade -q git+https://github.com/huggingface/transformers
!pip install torch

Load model via 🤗 Transformers

from transformers import AutoProcessor, AutoModelForMaskGeneration

processor = AutoProcessor.from_pretrained("martintomov/InsectSAM")
model = AutoModelForMaskGeneration.from_pretrained("martintomov/InsectSAM")

Notebooks

Explore InsectSAM's capabilities and integration with GroundingDINO through three Jupyter notebooks available on the RB-IBDM GitHub page:

• InsectSAM.ipynb: Training process
• InsectSAM_GroundingDINO.ipynb: Enhanced segmentation performance with GroundingDINO
• InsectSAM_script.ipynb: Image processing script

GitHub: https://github.com/martintmv-git/RB-IBDM/tree/main/InsectSAM