Industrial Autonomy
SECTOR 02

Industrial Autonomy

Defect data at production speed.

Visual inspection on the production line requires models trained on precisely labeled defect data. cvPal helps quality teams curate and maintain the datasets that power real-time defect detection — from surface scratches to assembly misalignment.

1

Consistent defect taxonomies

Standardize defect class names across plants and shifts. Replace 'scratch_type_a' with 'surface_scratch' globally. Remove deprecated classes. Split broad categories into granular sub-types. Label operations work across the entire dataset — no manual image-by-image editing.

2

Augment for production conditions

Factory lighting changes, camera angles shift, products rotate on conveyors. Augment with rotation, brightness variation, and flips to build models that work under real production variability. Annotation coordinates transform automatically.

3

Merge data from multiple lines

Combine defect datasets from different production lines, factories, or camera systems. Class reindexing happens automatically, deduplication catches overlapping images, and the merged result maintains split structure and version history.

4

Version-controlled retraining

When new defect types appear, add them to the dataset and create a new version. Export the updated dataset, retrain, and compare. If accuracy drops, restore the previous version and investigate. The full timeline is always available.