MycoMatch
How it works
MycoMatch helps identify fungi from microscope images. An analyst uploads a specimen image, and the system detects the key structures, finds the closest matches in a reference case library, and explains why they match.
Our Approach
The goal was a tool an expert can check, not a black box. Every candidate comes with an annotated specimen, a similarity score, citations to reference-book pages and a written rationale, and a doctor accepts, flags or rejects every result.
Computer Vision Pipeline
Images first pass an image-quality check, then a YOLOv8 detector finds the morphological structures. DINOv2 embeddings are searched with a FAISS index against the reference library to rank candidate matches.
Explainable Results
A language model (Gemini or Groq) turns the structured observations into a field-by-field explanation. Low-confidence matches are flagged as inconclusive rather than forced.
A Library That Learns
Accepted identifications can be saved back as new cases, and the index is rebuilt incrementally, so the library grows with use. Results export as a PDF report.
Desktop App With Accounts
A React desktop front end talks to a local Python API, with user accounts, an admin panel and activity logs.
Built with: Python, FastAPI, PyTorch, YOLOv8, DINOv2, FAISS, Gemini, Groq, React, Vite, Electron.
