Why “curate first, annotate smarter” is changing the shape of computer vision development

Automate quality gates

Replace subjective manual reviews with deterministic quality gates. Automated checks are the only way to detect systematic errors such as schema violations and class imbalances that are inevitably missed by human reviewers at scale.

from fiftyone import ViewField as F
# Find bounding boxes that are impossibly small
tiny_boxes = dataset.filter_labels(
    "ground_truth",
    (F("bounding_box")[2] * F("bounding_box")[3]) < 0.01
)

# Find samples where the model disagrees with ground truth
possible_errors = dataset.match(F("mistakenness") > 0.8)

# Schema Validation: Find detections missing required attributes
incomplete_labels = dataset.filter_labels(
    "ground_truth",
    F("occluded") == None
)

Preserve the provenance of annotations

Track curation decisions and annotation metadata to support iterative improvements. This provenance enables advanced analysis of which curation strategies result in the best model improvement, supporting continuous workflow optimization.

# Grab the "most unique" sample from a curated view of unique smaples
most_confusing_sample = unique_view.first()

# Add sample-level provenance
most_confusing_sample.tags.append("curated_for_review")

# Set metadata on the specific 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()

An integrated platform for curation-driven workflows

FiftyOne, Voxel51’s flagship open source computer vision platform, provides the tools you need to curate, annotate, and evaluate AI models. Provides a unified interface for data selection, QA, and iteration.

Source link