deep-danbooru
Use DeepDanbooru for anime-style image tag estimation, project and Danbooru-style dataset preparation, TensorFlow training, image evaluation, TFLite conversion, and experimental Grad-CAM workflows.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- deep-danbooru
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use DeepDanbooru for anime-style image tag estimation, project and Danbooru-style dataset preparation, TensorFlow training, image evaluation, TFLite conversion, and experimental Grad-CAM workflows.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/install-and-cli.md2.8 KB
- 参考文档references/repo-provenance.md1.6 KB
- 参考文档references/repo-routing-metadata.json332 B
- 参考文档references/troubleshooting.md3.6 KB
03 SKILL.md 正文
DeepDanbooru
Use this repo skill when a task names DeepDanbooru or deepdanbooru, asks for
anime-style multi-label image tagging, or needs the package's project, dataset,
training, evaluation, TensorFlow Lite, or Grad-CAM workflows. This skill covers
the 1.0.0 source contract represented by the provenance snapshot; read it before
refreshing against a changed checkout.
Route by task
- Project, tags, or dataset layout: read
project-data-setup. It ownscreate-project,download-tags,make-training-database,project.json,tags.txt, SQLiteposts, andimages/<prefix>/validation. - Training or checkpoint recovery: read
model-training. It owns config validation, supported ResNet/optimizer/loss values, data preflight,train-project, checkpoints, and.kerasexports. - Tag inference or evaluation: read
inference-evaluation. It ownsevaluate,evaluate-project, Python inference APIs, preprocessing, thresholds, recursive folders, and.txtsidecars. - Export or attribution artifacts: read
post-training-tools. It ownsconv2tflite, TFLite checks, and experimentalgrad-camoutput.
Use the owning sub-skill rather than trying to make the root a full API manual. The root references below contain cross-cutting install, CLI, and failure information.
Install and verify
The source package declares a base install plus a TensorFlow extra. For the repository's documented full workflow, install the TensorFlow requirements in a fresh compatible Python environment:
python -m pip install "deepdanbooru[tensorflow]"
# Or, when the distribution's published requirements file is available:
python -m pip install -r requirements.txt
The package imports TensorFlow and TensorFlow I/O through its public module and CLI paths. Do not call a CPU import proof of CUDA readiness. Start with the bundled no-download diagnostic:
python scripts/environment_smoke.py
python -m deepdanbooru --help
python -m deepdanbooru evaluate --help
The console entry point is deepdanbooru. CPU TensorFlow is the verified
correctness backend for this skill. evaluate is CPU-first unless
--allow-gpu is explicitly supplied; this source snapshot did not verify GPU
libraries or GPU performance.
Operating boundaries
- Do not download Danbooru tags, model weights, or datasets without explicit network/credential approval. The bundled scripts are offline and safe by default.
- Do not launch full training as a smoke test. Validate the project, tags, SQLite schema, image paths, and package imports first; training is expensive and writes checkpoints/exports.
- Treat the known 1.0.0
evaluate-projectloader defect as real: its native route calls the absentdd.data.load_tags_from_projectsymbol. For production folder inference, useevaluate --project-path ... --allow-folder, which uses the available project tag loader. The bundled project smoke helper exercises that fallback and does not claim an unverified native run. - The native
evaluate --save-txtpath can fail for an empty selected-tag list and overwrites an existing sibling sidecar withw. Use the skill helpersub-skills/inference-evaluation/scripts/save_txt_guard.pywhen sidecar output needs a local safety boundary: it rejects empty selections and sibling directories, and requires an explicit flag to overwrite a regular file. - Read
repo-provenance.mdbefore deciding whether this graph is stale; package/source version drift remains possible. - Read
install-and-cli.mdfor the full command map and dependency/version caveat. - Read
troubleshooting.mdfor package-wide recovery.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.