noe/arthur
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Port step 6: detect real footage with local MediaPipe

Browse source
This commit is contained in:
Olive Vaughn 2026-09-27 19:13:20 -04:00
parent 8a06835895
commit 32683efccf
27 changed files with 18258 additions and 54 deletions
Download patch file Download diff file Expand all files Collapse all files

17
frontend/public/mediapipe/README.md Normal file
View file

@ -0,0 +1,17 @@
# Local MediaPipe assets
`vision_bundle.js` and the four wasm loader/binary files under `wasm/` come
from `@mediapipe/tasks-vision` **1.0.1**, pinned in `frontend/package.json`.
`FilesetResolver.forVisionTasks` uses the SIMD pair where supported and the
no-SIMD pair elsewhere. The package and these files are Apache-2.0; see
[LICENSE](LICENSE).
`face_landmarker.task` is the official [Face Landmarker model](https://storage.googleapis.com/mediapipe-models/face_landmarker/face_landmarker/float16/1/face_landmarker.task).
Its SHA-256 is
`64184e229b263107bc2b804c6625db1341ff2bb731874b0bcc2fe6544e0bc9ff`.
These are served from `/mediapipe/` so detection needs no CDN at runtime. The
browser bundle is loaded as a script before the CLJS app because Shadow CLJS
cannot parse the package's CommonJS bundle (its dynamic `import()` is unsupported
by the current compiler). `flow/detect.cljs` is the sole call site for its
`Vision` global.