---

title: filter_generated() description: Python API reference for ciagen.filter_generated() : filter synthetic images by point-to-distribution quality scores. keywords: api, filter, python, quality filtering

---

filter_generated()

Filter generated images based on point-to-distribution quality scores.

from ciagen import filter_generated

kept = filter_generated(
    generated="data/generated/",
    method="top-k",
    value=100,
)

Parameters

Parameter	Type	Default	Description
generated	str \| Path	required	Directory containing generated images and metadata.yaml
ptd_scores	dict \| None	None	Pre-computed PTD scores (reads from metadata.yaml if None)
method	str	"top-k"	Filtering method: threshold, top-k, top-p
value	float	1000	Filtering threshold (interpretation depends on method)
metric	str	"mld"	PTD metric name to filter on
feature_extractor	str	"vit"	Feature extractor name whose scores to use

Filtering Methods

Method	value meaning
top-k	Keep the k images with smallest distances
top-p	Keep the top proportion (0 ≤ p ≤ 1) of images
threshold	Keep images with distance ≤ value

Returns

dict mapping metric names to feature extractor results:

{
    "mld": {
        "vit": {
            "/path/to/best_image.png": 0.45,
            "/path/to/second_best.png": 0.52,
        }
    }
}

Raises

• FileNotFoundError : if no metadata.yaml exists in the generated directory and ptd_scores is None
• ValueError : if method is not one of the valid options or value is out of range