π End-to-end encrypted cloud for everything.
Dart 2faandroidauthye2eeencryptionend-to-end-encryptionfluttergolanggoogle-photosgoogle-photos-alternativeiosopensourcephotosprivacyprivacy-toolstypescriptzero-knowledge
use crate::{
error::{MlError, MlResult},
models::Model,
onnx,
postprocess::l2_normalize,
runtime::MlRuntimeView,
types::FaceResult,
};
use super::FACE_INPUT_SIZE;
const FACE_INPUT_CHANNELS: i64 = 3;
pub(crate) fn run_face_embedding(
runtime: &MlRuntimeView<'_>,
aligned_faces: Vec<Vec<f32>>,
face_results: &mut [FaceResult],
) -> MlResult<()> {
if aligned_faces.is_empty() {
return Ok(());
}
if aligned_faces.len() != face_results.len() {
return Err(MlError::Postprocess(format!(
"aligned faces count ({}) does not match face result count ({})",
aligned_faces.len(),
face_results.len()
)));
}
let face_input_size = FACE_INPUT_SIZE as i64;
let expected_input_len = (face_input_size * face_input_size * FACE_INPUT_CHANNELS) as usize;
for (aligned, face_result) in aligned_faces.into_iter().zip(face_results.iter_mut()) {
if aligned.len() != expected_input_len {
return Err(MlError::Preprocess(format!(
"aligned face tensor length {} does not match expected {}",
aligned.len(),
expected_input_len
)));
}
let input = onnx::PreparedF32Input::new(aligned);
let (shape, mut embedding) = runtime.run(Model::FaceEmbedding, |session| {
onnx::run_f32(
session,
&input,
[1, face_input_size, face_input_size, FACE_INPUT_CHANNELS],
)
})?;
if shape.first() != Some(&1) || embedding.is_empty() {
return Err(MlError::Postprocess(format!(
"invalid single-face embedding tensor shape {:?} for data length {}",
shape,
embedding.len()
)));
}
l2_normalize(&mut embedding, f32::EPSILON);
face_result.embedding = embedding;
}
Ok(())
}