Automate high quality gates
Substitute subjective guide evaluation with deterministic high quality gates. Automated checks are the one strategy to catch systematic errors like schema violations or class imbalance that human reviewers inevitably miss at scale.
from fiftyone import ViewField as F
# Discover bounding bins which are impossibly small
tiny_boxes = dataset.filter_labels(
"ground_truth",
(F("bounding_box")[2] * F("bounding_box")[3]) < 0.01
)
# Discover samples the place the mannequin disagrees with floor fact
possible_errors = dataset.match(F("mistakenness") > 0.8)
# Schema Validation: Discover detections lacking required attributes
incomplete_labels = dataset.filter_labels(
"ground_truth",
F("occluded") == None
)
Preserve annotation provenance
Observe curation choices and annotation metadata to assist iterative enchancment. This provenance permits refined evaluation of which curation methods yield the most effective mannequin enhancements and helps steady workflow optimization.
# Seize the "most original" pattern from a curated view of distinctive smaples
most_confusing_sample = unique_view.first()
# Add sample-level provenance
most_confusing_sample.tags.append("curated_for_review")
# Set metadata on the precise labels (detections)
if most_confusing_sample.detections:
for det in most_confusing_sample.detections.detections:
det["annotator"] = "expert_reviewer"
det["review_status"] = "validated"
most_confusing_sample.save()
A unified platform for curation-driven workflows
Voxel51’s flagship open supply laptop imaginative and prescient platform, FiftyOne, supplies the mandatory instruments to curate, annotate, and consider AI fashions. It supplies a unified interface for information choice, QA, and iteration.
