Multimodal annotation
Check annotation quality.
Across five formats.
Inspect the source. Mark what matters. Validate the submission and review an independent grade. Forge connects each observation to a structured result.
Video
Review actions, object tracks and timing.
Model results attachedImage
Locate objects and check their labels and geometry.
Model results attachedText
Identify labels and spans in the source text.
Model results attachedStructured data
Check annotations across fields and records.
Model results attachedAudio
Mark sound events and their time intervals.
Model results attachedReview and interchange
Bring structure to the review.
Validate annotation data, compare two sets and inspect disagreement. Supported image and video boxes can move through CVAT XML 1.1 and Label Studio JSON.
Unsupported shapes and interpolation are rejected rather than silently changed. These are tested format subsets, not full compatibility with every feature of those tools.
Independent checks
Know what the grade means.
Certification tests the grader with reference work, empty submissions, negative controls and adversarial probes. Model campaigns then measure actual performance on the fixed tasks.
Reports expose outcomes and evidence digests while keeping reference answers and reusable task materials private.
How grading works →