CoverageCV · Merge with context

C·CoverageCV

DATASET MERGE LAB

Local workbench

YOUR PROJECTS

CoverageCVMake missing labels explicit.

●

Demo / Merge demo

Queue idle

BRING YOUR OWN DATA

Import a dataset

Package your COCO files, images and a coverage.json policy in one ZIP. Compilation checks the data before training can start.

Project nameChoose a dataset ZIPUp to 100 MB compressed · 500 MB unpacked

Or import an existing local spec

Absolute path to coverage spec

No dataset ready? Load the prepared chess project and explore real annotations.

CONTROLLED COMPARISON

Run an experiment

Compare supervision policies from one initialization, with the same images and update budget. Results use complete validation labels.

Updates per runRandom seedBatch sizeTime limit per run

Execution device

Training recipe

Ordinary training · controlStandard loss on the incomplete annotationsCoverageCV · our methodMissing labels are not assumed to be backgroundFully labeled referenceSame model with all labels; not a frontier model

This queue uses the selected device on your local machine. Jobs survive a workbench restart, and you can cancel them. No cloud workloads are launched.

Annotation coverage

Source evidence & identity

CoverageCV walkthrough

Recorded interface · captions on · no audio