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:
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¶
4. Generate Embeddings¶
After training, generate embeddings for analysis:
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¶
- Data Types: Learn about different data formats
- SimCLR Configuration: Customize training parameters
- Evaluation Tools: Explore analysis options
- Examples: See detailed use cases