blinklinmult.assets¶

Small files that ship with the package.

Only the demo clip lives here, and only because a demo with nothing to run is not a demo: a fresh pip install should be able to show the pipeline working without the user first obtaining a corpus.

The clip is the first 10 seconds of TalkingFace, 1.6 MB rather than the 22 MB source, carrying three annotated blinks (frames 168, 225 and 274) and its matching .tag slice so a prediction can be read against the annotation.

Nothing else belongs here. Model weights are downloaded from the Hub on first use and cached; the corpora are gigabytes and stay out of the wheel entirely.

Attributes¶

`ASSETS_DIR`	Directory this module lives in, and the files sit beside it.
`EXAMPLE_VIDEO`	TalkingFace, first 10 s at 30 fps.
`EXAMPLE_TAG`	Its annotation, sliced to the same 300 frames.

Package Contents¶

blinklinmult.assets.ASSETS_DIR¶: Directory this module lives in, and the files sit beside it.

blinklinmult.assets.EXAMPLE_VIDEO¶

TalkingFace, first 10 s at 30 fps.

Type:: The bundled demo clip

blinklinmult.assets.EXAMPLE_TAG¶

Its annotation, sliced to the same 300 frames.

Frame ids are unchanged from the source, so a prediction on this clip lines up with the published TalkingFace numbers.