🏋️ Training Your Model on FungiTastic

This page gives practical steps to get started training a model for any FungiTastic benchmark.

---

1. Setup

• Clone the repo and install requirements:

  git clone https://github.com/bohemianvra/FungiTastic.git
  cd FungiTastic
  pip install -r requirements.txt
  

• Download the dataset (see Downloading the Dataset).

---

2. Loading the Data

• Each benchmark split is a CSV/TSV file listing image paths and metadata.

• Example (Python/pandas):

  import pandas as pd
  
  df = pd.read_csv("fungitastic/closed_set_train.csv")
  print(df.columns)  # See available metadata fields
  

• Images are typically stored as paths in the DataFrame. Load with your favorite library (PIL, OpenCV, etc).

---

3. Training Scripts & Examples

• Example scripts for closed-set, few-shot, open-set, and segmentation are in examples/ in the repo.
  • E.g., examples/train_closed_set.py
• All baselines use standard PyTorch; configs and hyperparameters are in the paper and README.

---

4. Tips

• Always start with the Mini subset for fast prototyping.
• For multi-modal training (images + metadata), see the relevant examples.
• Use pre-trained weights as strong starting points—see Baselines & Models.

---

5. Extending to New Benchmarks

• FungiTastic is benchmark-agnostic: write your own splits, add modalities, or propose new tasks.

---

Next steps:
- Evaluation Protocols
- Baselines & Models