Skip to content

docs: getting started, overview, command line, library API - #13

Closed
201815054 wants to merge 3 commits into
mainfrom
docs/library-guides
Closed

docs: getting started, overview, command line, library API#13
201815054 wants to merge 3 commits into
mainfrom
docs/library-guides

Conversation

@201815054

Copy link
Copy Markdown
Collaborator

Four pages under docs/, and a one-line index in the README.

Page For a reader who is Lines
getting-started running something for the first time 100
overview asking what this library is and whether it fits 110
using-the-cli looking up a command or an option 158
using-the-library putting a model inside their own program 124

Why

The repository had a README and a model implementation guide, and nothing in between. There was no page for someone who has never run anything, no reference for the CLI options, and nothing describing how to call the models from an application — which is the ordinary way to use the library.

Pages are separated by what the reader is doing rather than by subsystem: learning, doing, looking something up, understanding. A tutorial that stops to explain the design, or a reference that turns into a walkthrough, serves neither reader.

getting-started uses BiRefNet rather than the README's MobileSAM example, because it needs no prompt coordinates to explain and produces a result you can look at in one command.

Verification

  • Every internal link and anchor resolves; every referenced repository path exists.
  • The getting-started commands were run end to end as written against a build of this branch.
  • Signatures in the reference sections were read out of postproc.h, tracker.h and vision.h.

No source files are touched. Independent of #11 and #12.

🤖 Generated with Claude Code

Adds four pages under docs/, and a one-line index in the README.

  getting-started    first run in five minutes, from the release package
  overview           what the library is; structure as code, weights as a file
  using-the-cli      every vision-cli command and option, one section per task
  using-the-library  the load/compute pattern, per-model entry points, bindings

The repository had a README and a model implementation guide, and nothing in
between: no page for someone who has never run anything, no reference for the
CLI options, and nothing showing how to call the models from an application.

Pages are separated by what the reader is doing -- learning, doing, looking
something up, understanding -- rather than by subsystem.

Every internal link and anchor resolves, every referenced path exists, and the
getting-started commands were run as written. No source files are touched.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
eunchae and others added 2 commits August 7, 2026 11:04
Removes bold and italic markers from the prose in the four guides. Structure --
headings, tables, code, links, lists -- is untouched; only the markers a reader
sees as noise when looking at the source are gone.

README.md is upstream's and keeps its own conventions.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The previous pass worked line by line, so it missed emphasis spanning two
lines and the asterisk form of italics.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@201815054

201815054 commented Aug 7, 2026

Copy link
Copy Markdown
Collaborator Author

Closing while a larger issue is investigated: with a trained checkpoint the pipeline does not reproduce the reference detections. Reproduced on main, so it predates these changes, but there is no point reviewing documentation for a flow whose end-to-end correctness is in question.

@201815054 201815054 closed this Aug 7, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant