lightly
Use LightlySSL for self-supervised computer-vision training, model components, CLIs, embeddings, data layout, evaluation, and repository maintenance.
- License 许可
- MIT
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- lightly
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use LightlySSL for self-supervised computer-vision training, model components, CLIs, embeddings, data layout, evaluation, and repository maintenance.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/package-overview.md2.5 KB
- 参考文档references/repo-provenance.md1.7 KB
- 参考文档references/repo-routing-metadata.json514 B
- 参考文档references/troubleshooting.md2.8 KB
03 SKILL.md 正文
LightlySSL Repo Skill
Use this skill when a task names lightly, LightlySSL, self-supervised computer vision, Lightly's CLI commands, SSL transforms/losses/heads, image-folder embeddings, YOLO crop helpers, or maintenance of the LightlySSL repository.
Quick setup
- Public install:
pip install lightly - Optional TIMM/ViT modules:
pip install "lightly[timm]" - Optional direct video support:
pip install "lightly[video]" - Minimal import check:
python - <<'PY'
import lightly
from lightly.loss import NTXentLoss
from lightly.transforms import SimCLRTransform
print(lightly.__version__, NTXentLoss, SimCLRTransform)
PY
For a richer environment smoke, run the bundled environment checker. It imports core modules, reports optional TIMM/video availability, and can do a tiny tensor check without downloads.
Route map
| Need | Read |
|---|---|
| Assemble datasets, collates, transforms, losses, projection heads, memory banks, or method-to-component mappings. | ssl-building-blocks |
| Adapt a bare PyTorch, PyTorch Lightning, or distributed Lightly SSL training recipe without downloading repo examples. | training-workflows |
Build or validate lightly-version, lightly-ssl-train, lightly-embed, lightly-magic, lightly-crop, Hydra overrides, input folders, embeddings CSVs, or YOLO crop labels. |
cli-data-embedding |
| Use KNN/linear evaluation utilities or choose repository checks for changes to losses, transforms, examples, docs, notebooks, or distributed tests. | evaluation-maintenance |
Shared references
- Package overview — public surfaces, optional extras, and workflow boundaries.
- Troubleshooting — cross-cutting install/import/backend/data/config problems before routing deeper.
- Repository provenance — source snapshot and refresh triggers.
- Router metadata — structured scenario placement used by the managed repo-skills router.
Operating rules
- Do not run training, embedding, crop, notebook, benchmark, or distributed commands until the user provides data paths, output policy, and runtime budget.
- Prefer bundled scripts for safe preflight: environment checks, component smoke checks, CLI command construction, folder validation, tiny YOLO fixtures, and synthetic SimCLR steps.
- Treat GPU, distributed, TIMM, and video support as optional unless the user explicitly asks for them; verify the relevant backend/extra before claiming it works.
- Keep source-code maintenance guidance separate from package-use guidance. When editing the repository, use the maintenance sub-skill to select focused checks before broad
make all-checks.
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.