Skip to main content
Collecting and labeling enough real-world images—especially rare surface defects, edge-case lighting, or new material finishes—is one of the biggest bottlenecks in industrial computer vision. The Repli5 platform is built for ML, perception, and quality-inspection teams who need to close those data gaps without weeks of manual collection or hand-labeling. Starting from your existing part photos, you can synthesize localized physical defects or generate material, color, and lighting variations using natural-language descriptions, reference close-ups, and reusable project templates—complete with automatic COCO segmentation masks, 2D bounding boxes, and automated visual quality evaluation.

​
Core Workflows & Capabilities

Defect Generation

Synthesize realistic, localized surface defects (scratches, dents, tears, wrinkles) onto defect-free (OK) part images with automatic COCO masks and 2D bounding boxes.

Variation Generator

Transform object colors, surface materials, and lighting in close-up photos while preserving existing geometry, corner wear, and surface defects.

Projects & Data Library

Upload .zip datasets, organize shared team workspaces with Projects, filter datapoints by custom tags, and export ready-to-train archives.

Quality Evaluation & Review

Automatically score every generated image across workflow-specific visual rubrics, review outputs, and regenerate individual images in-place.

​
Ready to Begin?

Quick Start Guide

Follow the step-by-step walkthrough to upload your first dataset, set up a project, choose the right workflow, and export your augmented training data.