yolozu 4.1.0

Interface-contract-first evaluation toolkit for mitigating catastrophic forgetting and domain shift via reproducible workflows.

YOLOZU (萬)

Japanese: Readme_jp.md | Chinese: Readme_zh.md

YOLOZU is an Apache-2.0 vision evaluation toolkit for teams that do not want workflow lock-in.

Bring your own inference. Export once. Evaluate fairly.

YOLOZU uses one stable predictions interface contract: wrapped predictions.json with protocol-pinned meta.export_settings.

1-Minute Demo

python3 -m pip install -U yolozu
yolozu demo overview

Writes demo_output/overview/<utc>/demo_overview_report.json.

flowchart LR
    A["Ultralytics"] --> D["wrapped predictions.json"]
    B["RT-DETR"] --> D
    C["Detectron2 / MMDetection / custom"] --> D
    D --> E["validate"]
    E --> F["evaluate"]
    F --> G["comparable report"]

Read These First

• docs/README.md: top-level docs map and shortest working paths
• docs/predictions_schema.md: the predictions interface contract
• docs/install.md: install, doctor, and environment setup

Primary Focus

• Main lane: evaluate precomputed predictions fairly across frameworks and runtimes
• Secondary lane: export/train scaffolds that feed the same predictions interface contract
• Advanced lane: continual learning, TTT, SynthGen, and backend parity research paths

Capability Maturity

• Stable: prediction validation/evaluation, wrapped predictions.json, repo smoke/demo path, install/doctor flow
• Experimental: backend parity, benchmark orchestration, SynthGen intake and handoff, macOS/MPS evaluation paths
• Research: continual learning, self-distillation, TTT, Hessian refinement

Production Readiness

• Production-ready today: prediction validation/evaluation and the predictions interface contract
• Needs qualification in your environment: backend parity, benchmark orchestration, SynthGen handoff, macOS/MPS paths
• Research-oriented: continual learning, self-distillation, TTT, Hessian refinement
• Full details: docs/production_readiness.md

Who This Is For

• You already have predictions and want fair cross-framework evaluation.
• You want an Apache-2.0 evaluation layer without rewriting your training stack.
• You do not want framework-native evaluation differences to become silent metric drift.

Not The Best Fit

• You want one end-to-end training framework with one-click defaults.
• You do not need cross-framework comparison or a stable predictions interface contract.

Why Not Just Use Framework-Native Evaluation?

Framework-native evaluation is convenient inside one stack, but it is harder to compare fairly across stacks. YOLOZU keeps the evaluation boundary at one predictions interface contract so the comparison path stays pinned even when the inference stack changes.

Where To Go Next

• Evaluate precomputed predictions: docs/external_inference.md
• Train, export, then evaluate: docs/training_inference_export.md
• Compare backends and benchmark paths: docs/backend_parity_matrix.md, docs/benchmark_mode.md
• Prepare YOLOZU-synthgen handoff: docs/synthgen_repo_integration.md
• Tool and manifest references: docs/tools_index.md, tools/manifest.json

More Than The Demo

• Advanced docs map: docs/README.md
• Real-image showcase: docs/assets/readme_multitask_showcase.png
• Learning and research workflows: docs/learning_features.md

Repo Users

python3 -m pip install -e .
bash scripts/smoke.sh

More repo-first guidance:

• Docs index: docs/README.md
• Install details: docs/install.md
• Manual sources: manual/README.md

Support And Legal

• Support: docs/support.md
• License policy: docs/license_policy.md
• Apache-2.0 license: LICENSE
• Latest release: GitHub Releases
• Zenodo software DOI: 10.5281/zenodo.18744756
• Zenodo manual DOI: 10.5281/zenodo.18744926