Evaluation & Visualization Overview

The evaluation module provides comprehensive tools for analyzing and visualizing embeddings generated by the SimCLR framework. This module helps researchers understand the quality and biological relevance of learned representations.

Core Functionality

Embedding Analysis

  • Dimensionality Reduction: PCA analysis with variance explained
  • Clustering Quality: Silhouette scores

Visualization Tools

  • Scatter Plots: 2D projections with image thumbnails
  • UMAP/t-SNE: Non-linear dimensionality reduction

Input Requirements

The evaluation tools expect:

CSV File Format

filename,x1,x2,x3,...,x1024
bird_sex_001.jpg,0.12,-0.45,0.78,...,0.23
bird_sex_002.jpg,-0.34,0.67,-0.12,...,0.89

Required Columns

  • filename: Path to original image
  • Embeddings: Feature vectors (x1, x2, ..., xN)

Usage Example

python eval/eval_silhouette_variance.py \
  --embeddings embeddings.csv \
  --output results/ \

Output Files

The evaluation generates: - Statistical summaries: CSV files with metrics

Next Steps