v1.4.4

Modularity, viewer integration, and JOSS resubmission release, responding to pre-review feedback on openjournals/joss-reviews#10182 and #10183.

Added

• Bundled viewer: The FastAPI service now serves Video Annotation Viewer's static build at /viewer, same-origin and zero-config (VIDEOANNOTATOR_ENABLE_VIEWER to disable). VAV remains fully usable standalone.
• Contract test: Added tests/contract/test_viewer_contract.py, which validates VideoAnnotator's exporter output against VAV's actual Zod schemas.
• Roadmap docs: Added docs/development/roadmap_v1.6.0.md (plugin ecosystem, local LLM/Ollama backend) and rewrote roadmap_v1.5.0.md around the modularity spec (specs/003-modular-pipeline-architecture).

Fixed

• Person-tracking contract gap: COCO output was missing person_id/person_label/label_confidence/labeling_method when identity labeling was disabled, which VAV requires as non-optional.
• setup-db DetachedInstanceError on the first command in the README quickstart.
• Hidden CLI commands: validate-emotion, generate-token, and setup-db were missing from --help due to a misplaced __main__ guard.
• First-run friction: torch/CUDA warmup moved off the first health-check request into a startup task; added a one-click /viewer-connect login bridge.
• Silent face-pipeline fallback: the face pipeline alias could silently pick an unconfigured/experimental variant instead of erroring; OpenFace3 now fails fast with one clear error instead of one per frame.
• Windows SQLite teardown (openjournals/joss-reviews#10182): the engine-disposal fix from this release didn't cover two test fixtures that construct SQLiteStorageBackend directly; both now close the backend in teardown.

Documentation

• Full stale-link/version sanity pass ahead of JOSS resubmission: fixed ~14 stale src/* paths in reviewer-facing docs left over from the src/videoannotator/ package restructure, backfilled missing v1.4.3/v1.4.4 changelog entries, corrected pyproject.toml mypy config pointing at removed paths, and fixed several stale version pins (INSTALLATION.md, Docker.md, output_naming_conventions.md, CITATION.cff).
• Added the interface figure to the combined JOSS paper and documented the /viewer integration and contract test as evidence of coordinated maintenance between VideoAnnotator and VAV.

See CHANGELOG.md for the full itemized history, and PR #6 for the bulk of this release's changes.

v1.4.3

Patch release fixing installability and CI raised during JOSS review (openjournals/joss-reviews#10182).

Fixed

• Restored missing storage modules (storage/manager.py, storage/providers/*) that an unanchored storage/ rule in .gitignore had silently excluded from the repository. Fresh clones previously failed with No module named videoannotator.storage.manager, and the API/worker could not start.
• Cross-platform install: pinned the torch/torchvision/torchaudio trio to the matched 2.6.0 release and restricted the CUDA (cu124) index to Linux, so macOS/Windows resolve the CPU build from PyPI. Previously uv sync failed on any non-Linux/non-CUDA platform.
• Bumped openai-whisper from ==20240930 to >=20250625.
• Repointed reviewer documentation from the removed scripts/verify_installation.py to the built-in videoannotator diagnose command.
• Cleared pre-existing lint, type-check, and test failures. CI is now green on Linux and macOS; the Windows leg is informational pending #5.

See PR #4 for the full set of changes.

v1.4.2 - JOSS Review release

This release accompanies the submission of VideoAnnotator to the Journal of Open Source Software. A parallel submission is being made for the companion project Video Annotation Viewer.

Below is a summary of all changes across the v1.4.x release series (v1.4.0 -- v1.4.2).

---

Added (v1.4.0)

• Flexible storage: New artifact download capabilities, including source video retrieval.
• Database-backed authentication: Migrated from file-based to database-backed auth for improved security and scalability.
• Artifacts API: New endpoint GET /api/v1/jobs/{id}/artifacts to download job results as a ZIP archive.
• Container tooling: Baked hadolint into Docker images and devcontainer; added git-lfs to CPU/GPU Dockerfiles.

Changed

• CLIP migration (v1.4.2): Migrated scene-classification pipeline from clip to open_clip, using the LAION-2B pretrained ViT-B-32 model for improved availability and reproducibility.
• HuggingFace auth (v1.4.2): Updated diarization and Whisper pipelines to use the current token parameter instead of the deprecated use_auth_token.
• JOSS manuscript (v1.4.1): Consolidated the paper into paper/paper.md and replaced docs/joss.md with a pointer to avoid divergence.
• Repository hygiene (v1.4.1): Moved top-level helper scripts into organized subfolders under scripts/; updated imports to the videoannotator.* namespace.
• Entrypoints (v1.4.1): api_server.py now acts as a compatibility wrapper; documentation recommends the videoannotator CLI.
• README (v1.4.1): Rationalized setup/install instructions, fixed broken links, replaced hard-coded test/coverage claims with CI status badges.
• Devcontainer (v1.4.2): Simplified forwarded-port list to the single default API port (18011).

Fixed

• Artifact downloads (v1.4.0): Ensured source video files are included in the downloaded artifact ZIP.
• Database GUID handling (v1.4.2): Added defensive error handling in the GUID type decorator for malformed UUID values.
• Diarization init (v1.4.2): Wrapped model loading in explicit error handling with a clear log message on failure.
• Docs (v1.4.1): Standardized examples on canonical API port 18011 and corrected Docker run port mappings.
• Docs (v1.4.1): Replaced placeholder docs/usage/accessing_results.md with a real results retrieval guide.

Removed (v1.4.2)

• Voice emotion baseline: Removed voice_emotion_baseline pipeline metadata and associated tests (superseded by LAION EmoNet voice pipeline).

Documentation (v1.4.2)

• Added JOSS cover letter (paper/cover_letter.md).
• Updated paper bibliography and CITATION.cff to v1.4.2.

---

Full Changelog: v1.3.0...v1.4.2

VideoAnnotator v1.3.0 - Production Reliability & Critical Fixes

Released: October 31, 2025 🎃

This release addresses critical production blockers identified during client integration testing and establishes a production-ready foundation for JOSS publication.

Highlights

Core Improvements

• Job Cancellation: Cancel running jobs via API with proper GPU cleanup and worker management
• Persistent Storage: Configurable STORAGE_DIR with retention policies - no more data loss on restart
• Config Validation: Schema-based validation catches errors before job submission
• Secure by Default: AUTH_REQUIRED=true with automatic API key generation and restricted CORS

Infrastructure

• Package Namespace: Migrated to modern videoannotator.* namespace (PEP 517/518)
• Enhanced Diagnostics: New CLI commands (diagnose system/gpu/storage/database/all)
• Environment Config: 19 configurable environment variables with full documentation

Quality & Testing

• 720/763 tests passing (94.4%) - improved from 607 (79.6%)
• Fixed 113 tests with real audio/video fixtures
• ffmpeg installed across all Docker images
• Exceeds 95% target by 23 tests

Bug Fixes

• Fixed pipeline import and name resolution failures
• Added missing pipeline metadata
• Resolved ephemeral storage data loss risks
• Standardized error formats across all API endpoints

Documentation

• 10+ new documentation files
• Complete API documentation with curl examples
• JOSS reviewer quick start guide
• Security configuration guide
• Migration guide with automated script

Success Criteria - ALL MET ✅

• Zero job failures due to pipeline naming
• Zero data loss on server restart
• All running jobs cancellable via API
• Invalid configs rejected at submission
• Authentication required by default
• 84% task completion (56/67 tasks)

See CHANGELOG.md for complete details.

What's Next

v1.4.0 (Q2 2026): First Public Release + JOSS Paper submission

v1.2.2 - import fixes

Minor maintenance release.

Normalized internal imports (removed obsolete src. prefixes) resolving a server startup error.
Updated CLI server target to api.main:app and bumped version metadata.
No API or CLI breaking changes.

VideoAnnotator 1.2.1 – Registry Expansion & Stabilization

Highlights

• YAML-driven pipeline registry with richer taxonomy (family, variant, tasks, modalities, capabilities, backends, stability).
• Auto‑generated pipeline spec + emotion output format and validator.
• Standard error envelope + enriched system health endpoint.
• New face & voice emotion pipelines (OpenFace3 embedding, LAION face/voice, baseline emotion).
• Light post-freeze stabilization: optional legacy API key validation and Windows-safe ASCII logging.

Added

• Dynamic pipeline metadata exposure (/api/v1/pipelines, CLI videoannotator pipelines --detailed --json).
• Emotion schema + validator + naming conventions spec.
• Health endpoint now reports pipeline count, uptime, and queue snapshot.
• New pipelines registered (face_openface3_embedding, face_laion_clip, voice_emotion_baseline).

Changed

• Replaced legacy single category field with multi-dimensional taxonomy arrays.
• All pipeline discovery now sourced from registry (no hard-coded lists).
• Logging: removed emojis in runtime output in favor of [OK], [WARNING], [ERROR].

Compatibility

• Non-breaking for existing clients except: invalid API keys now correctly rejected (401).
• Anonymous endpoints remain accessible where previously allowed.

Upgrade Notes

If you referenced category, map logic to tasks, pipeline_family, or modalities.
Use: videoannotator pipelines --json for machine-readable metadata.
Regenerate pipeline spec after adding pipelines via the provided script.

VideoAnnotator v1.2.0 - REST API, Docker support, uv packaging

Major release featuring:

• Complete REST API with FastAPI and integrated background processing
• Modern Docker deployment system with CPU, GPU, and dev containers
• uv package manager migration for 10x faster dependency management
• Ruff integration for modern linting and formatting
• Comprehensive deployment documentation and troubleshooting

Production-ready features:

• Real-time job processing and status tracking
• Interactive API documentation at /docs
• Cross-platform compatibility and container orchestration ready
• Secure authentication framework with JWT tokens
• Complete backward compatibility with existing pipelines

Technical improvements:

• Fixed Docker builds with proper package management
• Optimized container sizes and build times
• Enhanced error handling and recovery systems
• Modern development workflow with integrated tooling