Skip to main content

OpenProcessor: Vision and goals

The canonical text is docs/VISION_AND_GOALS.md in the repository. This page is a short summary. Read the repository file for the full scope and standards.

What it is for​

OpenProcessor is a generic backend for computer-vision dataset curation and training. It works for any image domain. What an item or a region is comes from data you own: a class registry, a region profile, a prompt pack and a VLM endpoint, stored per project. It is released as a matched pair with its frontend, Cropwright.

The pipeline, end to end: ingest, detection, region stage (one or many boxes per item), embedding and clustering, human and AI review, import of labeled datasets, export, training, bake-off, and promotion back into serving.

What v0.6.0 delivers​

  • Isolated projects, each with its own indexes, directories and class registry.
  • Editable prompt packs and region profiles, versioned and revertible.
  • One or many regions per item.
  • Per-project keyboard shortcuts.
  • A selectable VLM, local or remote.
  • Dataset import (YOLO, COCO, OpenProcessor exports) and project combine, with classes mapped by name.
  • A one-line installer.
  • Assist runs: send a crop, a selection, a cluster or a filter to the VLM and apply the proposed label explicitly.
  • Bulk ingest from a folder, a manifest or S3, with a preview, and measured ingest speed-ups (decode once, uint8 inputs, FP16 embeddings, GPU decode, multi-GPU embedding).
  • Documentation for every feature, and a public, reproducible cars-and-wheels example.

Standards the code is held to​

  • The backend stands on its own. It does not need the frontend to be correct or useful.
  • Review depth matches real risk.
  • No dead code, no compatibility shims, no routes left running next to their replacement.
  • Class identity is the name, never the raw index, at every boundary.
  • Fail closed: an unbound project, a stale registry or an ambiguous state refuses the operation.
  • Work is merged in waves, each with its own gate.

Not in v0.6.0​

The full zero-copy GPU pipeline (issue #40), the 1M-image scale step, request-driven processing plans, faces and OCR on resident crops, a technical white paper and more domain showcases are planned for the next release and after.

More detail​