---
license: apache-2.0
datasets:
- martintmv/rb-ibdm-l
---

# InsectSAM: Insect Segmentation and Monitoring

<p align="left">
  <a href="" rel="noopener">
 <img width=200px height=200px src="https://i.imgur.com/hjWgAN9.png alt="Project logo"></a>
</p>

## Overview

InsectSAM is an advanced machine learning model tailored for the DIOPSIS camera systems, which are dedicated to Insect Biodiversity Detection and Monitoring in the Netherlands. Built on Meta AI's `segment-anything` framework, InsectSAM excels at segmenting insects from complex backgrounds, enhancing the accuracy and efficiency of biodiversity monitoring efforts.

## Purpose

This model has been meticulously trained to identify and segment insects against a variety of backgrounds that might otherwise confuse traditional algorithms. It is specifically designed to adapt to future changes in background environments, ensuring its long-term utility in the DIOPSIS project.

## Model Architecture

InsectSAM utilizes the advanced capabilities of the `segment-anything` architecture, enhanced by our custom training on an insect-centric dataset. The model is further refined by integrating with GroundingDINO, improving its ability to distinguish fine details and subtle variations in insect appearances.

## Quick Start

### Prerequisites

- Python
- Hugging Face Transformers
- PyTorch

### Usage

#### Install 
``` bash
!pip install --upgrade -q git+https://github.com/huggingface/transformers
!pip install torch
```
#### Load model directly via HF Transformers 🤗
``` bash
from transformers import AutoProcessor, AutoModelForMaskGeneration

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

### Notebooks

Two Jupyter notebooks are provided to demonstrate the model's capabilities and its integration with GroundingDINO:

- **InsectSAM.ipynb**: Covers the training process, from data preparation to model evaluation.
- **InsectSAM_GroundingDINO.ipynb**: Demonstrates how InsectSAM is combined with GroundingDINO for enhanced segmentation performance.

Check out the notebooks on RB-IBDM's GitHub page - https://github.com/martintmv-git/RB-IBDM/tree/main/InsectSAM