IF Images Croissant Dataset + DenseNet Embeddings Demo

Generate image embeddings from Croissant datasets using DenseNet and package results in RO-Crate format.

What it does

1. Loads CM4AI immunofluorescence images from a Croissant dataset
2. Generates 1024-dimensional embeddings using pre-trained DenseNet121
3. Saves embeddings and metadata to TSV files
4. Creates an RO-Crate package for reproducible research

Installation

pip install mlcroissant torch torchvision Pillow pandas numpy tqdm fairscape-cli

Usage

1. Place cm4ai_if_images_croissant.json in the working directory
2. Run jupyter notebook demo.ipynb
3. Execute cells sequentially

Outputs

• ./densenet_embeddings/image_embeddings.tsv - 1024-dim embedding vectors
• ./densenet_embeddings/ro-crate-metadata.json - Research object metadata