Hi all, I've published a blog post on rapidly labeling camera trap data using ChatGPT for species identification and a simple object detection model to get the bounding boxes. With this setup, I could run through a set of camera trap recordings and automatically obtain 520 labeled images with detected animals, with only some minutes of manual work by controlling the output, and batch relabel samples of the same species that got different spellings from ChatGPT, e.g. "Alaska moose" and "Alaskan moose".



While presented in Edge Impulse, this approach can be replicated outside the platform using Python scripts for example. I’d love for you to give it a read and share your thoughts on its potential usefulness in real-world applications. 



(For now, I limited the complexity to only support images with one bounding box, which works pretty well in environments with mostly solitary species. I'll work on support for multiple animals, especially if there turns out to be a big interest around this!)



Link to blog: 

Read "Improving Camera Traps to Identify Unknown Species with GPT-4o" on Edge Impulse

Improving Camera Traps to Identify Unknown Species with GPT-4o

When deploying AI models for wildlife monitoring to remote cameras in the field, a common challenge is dealing with unexpected animal species that were not accounted for during training. Here's an approach to improve that.

Edge Impulse • 17 Jul 2024

Video: