Research-Based Insect Biodiversity Detection and Monitoring
ICT & Artificial Intelligence
Client company:ARISE, Naturalis Biodiversity Center
Project description
Our project is an open-source machine learning model specifically tailored for DIOPSIS camera systems and ARISE algorithms, aimed at enhancing Insect Biodiversity Detection and Monitoring in the Netherlands. By leveraging the Meta AI segment-anything framework, InsectSAM is fine-tuned to accurately segment insects from complex backgrounds, significantly improving the precision and efficiency of biodiversity monitoring algorithms. It is particularly effective in environments with diverse backgrounds that attract various insect species. Developed using Python, PyTorch, Hugging Face Transformers, and OpenCV, InsectSAM is designed for easy integration into existing DIOPSIS and ARISE workflows. This ensures a seamless experience for researchers and developers, facilitating advanced insect biodiversity studies and contributing to environmental conservation efforts.
Context
InsectSAM - Insect Detect ML is an open-source machine learning model tailored for the DIOPSIS camera systems and ARISE algorithms, dedicated to Insect Biodiversity Detection and Monitoring in the Netherlands.
Results
We successfully helped our clients achieve their goal of handling complex floral backgrounds and efficiently segmenting insects. This resulted in the creation of three machine learning models and two Hugging Face Space demos to showcase our work.
Demo 1 - Insect Model Zoo:
In this demo, you can compare the performance of InsectSAM, YOLO v8, and Detectron2 by selecting an image from the provided examples. Note that this demo currently runs on a CPU, so loading times for new images might be slower.
Demo 2 - InsectSAM Inference:
This demo offers a comprehensive testing environment for InsectSAM. Users can experiment with cropping bounding boxes from segmentation and detection results and obtain a JSON string with the insect locations. This demo runs on a free GPU, providing faster loading times and better performance.
Models:
We have included three model cards for InsectSAM, YOLO v8, and Detectron2. These models are stored and publicly available for everyone as we love contributing to open source.
These tools and demos illustrate the efficiency and precision of InsectSAM in tackling the challenges of insect biodiversity detection amidst complex backgrounds.
About the project group
We are three software engineers from Bulgaria with strong backgrounds in AI and Machine Learning. We work on innovative projects both at Fontys University and externally. Our group is dedicated, collaborative, and always striving to push the boundaries of technology.
