Skip to content

✂️ Segmentation Task

Overview

Accurate segmentation of fungal body parts (caps, gills, stems, etc.) is crucial for fine-grained recognition and interpretability.
FungiTastic–Mini includes part segmentation masks for a subset of species.

Use Cases

  • Automated species identification pipelines
  • Explainable AI and morphological analysis
  • Training and evaluating segmentation models on natural data

Data & Splits

  • FungiTastic–Mini: Images with expert-checked segmentation masks
  • Classes: cap, gills, pores, ring, stem (and background)
  • Split: train/val/test following Mini subset

Evaluation Protocol

  • Semantic Segmentation: mean Intersection over Union (mIoU)
  • Instance Segmentation: mean Average Precision (mAP)

Baselines & Results

  • Baseline: Zero-shot pipeline (GroundingDINO + SAM)
  • Baseline IoU: ~89% (but misses small/atypical parts)
  • Further improvements possible with domain-specific models

See Baselines & Models for details.

Quick Start

  • Download Mini subset and annotation masks
  • Use sample notebooks/scripts in usage/training.md