steps.clip.model.evaluator

evaluator

Functions:

Name	Description
evaluate	Evaluate a CLIP model on an image-only dataset using clustering.

evaluate(model, dataset)

Evaluate a CLIP model on an image-only dataset using clustering.

This step: - Loads the trained CLIP model and processor. - Runs inference on the dataset to extract image embeddings. - Applies UMAP to reduce dimensionality. - Uses DBSCAN to identify clusters in the embedding space. - Saves and logs clustering visualizations (UMAP plot, cluster image grids, outliers).

Parameters:

Name	Type	Description	Default
model	CLIPModel	The trained CLIP model to evaluate.	required
dataset	CocoDataset	The image dataset to evaluate the model on.	required