Modules
High-level modules in the product framework. Today reflects OneOpen ML Studio 1.0.0. Items marked Target are direction only — not shipped.
Module |
Role |
Today (1.0.0) |
|---|---|---|
Identity & access |
Orgs, users, roles, SSO, API tokens |
Local users + sessions + API tokens; Team/Enterprise auth via |
Project management |
Task type, schema, members, settings |
Projects, classes, workspace parent, project members / roles |
Data sources |
Folders, ZIP, cloud, DBs, sync |
Folder / ZIP / upload; JSONL/CSV for sample modalities |
Dataset registry |
Samples, schema, lineage, tags |
Project-scoped images + samples; basic lineage on versions |
Annotation workspace |
CV / NLP / audio / tabular / GenAI tools |
CV: boxes, polygons, classification. Sample UIs for text, tabular, audio, video, LLM |
Task management |
Assign, lock, submit, reassign |
Soft locks, assignments, review queue |
Review & QC |
Multi-stage review, IAA, golden sets |
Review queue (approve / reject / changes). Full IAA / golden sets = target |
Profiling |
Image/text/tabular/audio stats |
Image + sample profiles |
Quality engine |
Leakage, corrupt files, invalid labels |
Image + sample checks |
Data cleaning |
Dedup, empty drop, flag / leakage fixes |
Per-modality remediations |
Training readiness |
Score vs task/adapter/export |
CV + sample train gates |
Transformation |
Resize, encode, filter, embeddings |
Preprocess in dataset versions |
Augmentation |
CV / text / audio / tabular |
CV aug in versions |
Splitting |
Stratified, group, time, k-fold |
Random / stratified train/valid/test |
Versioning |
Immutable versions + lineage |
Numbered versions with config snapshot |
Data explorer |
Grid, filters, embeddings |
Image grid + sample lists |
Pipeline builder |
Visual / YAML workflows |
Visual DAG editor + YAML/JSON runner |
Training adapters |
Framework-neutral train API |
Ultralytics, LibreYOLO, Torchvision; sklearn; HF prepare; modality packaging |
Experiments |
Compare runs, metrics, artifacts |
Experiment list, compare, tags/notes |
Model registry |
Lifecycle Experimental → Production |
Register weights + lifecycle states |
Active learning |
Uncertainty → retrain loops |
Unlabeled / confidence / entropy / margin selection |
Notifications |
In-app, email, Slack, webhooks |
Target |
Audit |
Append-only security events |
Enterprise audit stubs when enabled |
Annotation modalities
Available today (visual / CV)
Task |
Label tools |
Notes |
|---|---|---|
Object detection |
Bounding boxes |
Primary path; YOLO Label Assist |
Instance / semantic segmentation |
Polygons (+ SAM → polygon) |
No pixel-mask brush yet |
Image classification |
Whole-image class |
YOLO-Cls / Torchvision catalogs |
Not shipped yet (target): keypoints, OBB, OCR / document, point cloud, video tracking timeline, true mask painting.
Other modalities today
Area |
Available |
Depth |
|---|---|---|
Text |
Classification, NER |
Label UI + sklearn / HF prepare |
Tabular |
Classification, regression |
Label UI + sklearn |
Audio / video |
Classification, transcription / action |
Import → curate → package (baselines / prepare-only) |
LLM |
Instruction, preference (DPO), RAG eval |
Dataset packaging — not in-process GPU fine-tune |
Roles
Builtin roles include Organization Owner/Admin, Workspace Admin, Project Manager, Annotator, Reviewer, ML Engineer, Viewer, and custom roles — with backend-enforced permissions in Team/Enterprise. Local mode uses a Personal org without mandatory login.