Quick Start

Get up and running with Phenoscape in just a few steps.

Basic RGB Analysis

1. Prepare Your Data

Organize your RGB images in a directory structure:

data/
├── image1.jpg
├── image2.jpg
└── ...

For the evaluation purposes, the images' names should be in the format species_sex_001.jpg, species_sex_002.jpg, etc. For training SimCLR and generating embeddings, the images' names could be in any format e.g., image1.jpg, image2.jpg, etc.

2. Configure Training

Create or modify config.yaml:

data_dir: "path/to/your/data"
out_dir: "path/to/your/output"
backbone: "resnet50"
weights: "DEFAULT"
lr: 0.001
batch_size: 32
max_epochs: 100

3. Train SimCLR Model

cd simclr
python scripts/simclr_birdcolour.py --config config.yaml

4. Generate Embeddings

After training, generate embeddings for analysis:

python scripts/generate_embeddings.py --model-path outputs/model.ckpt --data-dir path/to/data

5. Visualize Results

cd ../eval_vis
python evaluate.py \
  --csv_path embeddings.csv \
  --image_dir path/to/images \
  --output_dir visualizations \
  --labels species color \
  --emb_num 1024

Multispectral Analysis

For 7-channel multispectral data:

cd simclr
python scripts/simclr_kornia_spectral.py --config configs/config_kornia_multispectral.yaml

Hyperspectral Analysis

For 408-band hyperspectral data:

cd simclr
python scripts/simclr_birdcolour_kornia_hyperspectral.py --config configs/config_kornia_hyperspectral.yaml

Expected Outputs

After running the pipeline, you'll get:

  • Trained model: checkpoints/model.ckpt
  • Embeddings: CSV file with feature vectors
  • Visualizations: UMAP/t-SNE plots with image thumbnails
  • Metrics: Statistical analysis results

Next Steps