Roadmap

Shipped capabilities in OneOpen ML Studio 1.0.0, and what is still planned.

Shipped

Computer vision

  • Detection, instance/semantic segmentation, image classification

  • Boxes, polygons, SAM assist, YOLO Label Assist

  • Dataset profiling, quality engine, training readiness gate

  • Data cleaning remediations (dedup, empty drop, leakage / flag fixes) before version / train

  • Stratified / seeded splits; version snapshots with preprocess & augment

  • YOLO / COCO export; local Ultralytics, LibreYOLO, and Torchvision train backends

  • YOLO-Cls / YOLO-Seg plus ResNet, EfficientNet, ViT, Mask R-CNN, DeepLab, and related models

  • Plugin manager with on-demand install of training extras

Text, tabular, audio, video, LLM

  • Text classification & NER; tabular classification & regression

  • Audio classification & transcription; video classification & action recognition

  • Instruction tuning, preference (DPO), and RAG eval datasets

  • Sample import/export (CSV / JSONL); modality-aware annotate UIs

  • sklearn, Hugging Face prepare, and modality packaging adapters

Collaboration & ops

  • Local users, sessions, API tokens; orgs → workspaces → projects + memberships

  • Soft sample locks, annotation revisions, assignments, review queue

  • Experiment tracking and model registry lifecycle

  • Docker Compose (API, worker, Postgres, Redis, MinIO)

  • Celery training queue with in-process fallback; S3-compatible artifacts

  • Plugin registry and visual pipeline DAG editor + YAML/JSON runner

  • Active learning selection (unlabeled, confidence, entropy, margin)

  • Team / Enterprise first-time /setup wizard (ONEOPEN_DEPLOYMENT_MODE)

  • Kubernetes starter manifests; enterprise audit + SSO config stubs

Planned

  • Full SSO handshake (OIDC / SAML / LDAP)

  • Real-time multi-user cursors / WebSockets

  • Document / OCR project type

  • Deeper immutable dataset lineage

  • Shared GPU scheduling across workers

Guides