bioclip2

Information extraction pipeline that makes the Quogue Wildlife Refuge's 72k-image archive searchable: zero-shot species and scene tags embedded as IPTC/XMP keywords.

A modular, reproducible pipeline that tests whether biological images can be grouped by species without training labels. It extracts features using foundation models, runs multiple unsupervised clustering methods, and evaluates clustering quality using ground-truth species labels. It also helps to annotate the clusters using BioCLIP-2

A Python tool for estimating animal GPS coordinates from camera-trap imagery using known camera position, bearing, and pinhole geometric projection. Uses YOLO+SAHI sliced inference for detection and BioCLIP 2 (via open_clip) for zero-shot species classification.

Repository featuring all additional content for Luka Kowalchuk's MSc Thesis on: A Comparative Evaluation of Vision Transformer and Convolutional Neural Network Performance for Forensically Relevant Dipteran Classification

Multi-image, multi-label plant phenology classification from iNaturalist photos

Python utilities for nature observations processing - images and bioacoustics. Including MCP.

A modular, reproducible pipeline that tests whether biological images can be grouped by species without training labels. It extracts features using BioCLIP-2 image embeddings, runs multiple unsupervised clustering methods, and evaluates clustering quality using ground-truth species labels.