recognition

recognition

Recognition and FIQA inference helpers.

Classes

Name	Description
RecognitionRunner	Runs recognition and optional FIQA sessions for aligned face crops.

RecognitionRunner

recognition.RecognitionRunner(
    models,
    rec_inference_sessions,
    ort_fiqa=None,
    extended,
    concat_embeddings,
    image_size=(112, 112),
    keypoint_recognition_models=None,
    image_input_name=SEPAELV6_IMAGE_INPUT,
    keypoints_input_name=SEPAELV6_KEYPOINTS_INPUT,
)

Runs recognition and optional FIQA sessions for aligned face crops.

Methods

Name	Description
compute_batch	Compute embeddings and optional FIQA scores for aligned face crops.
compute_one	Compute embeddings and optional FIQA score for one aligned face.

compute_batch

recognition.RecognitionRunner.compute_batch(
    bgr_aligned_batch,
    aligned_keypoints_batch=None,
)

Compute embeddings and optional FIQA scores for aligned face crops.

compute_one

recognition.RecognitionRunner.compute_one(
    bgr_aligned_face,
    aligned_keypoints=None,
)

Compute embeddings and optional FIQA score for one aligned face.

Functions

Name	Description
looks_like_cuda_oom	Best-effort match for CUDA out-of-memory errors raised by ONNX Runtime.
try_compute_embeddings_batch	Call a batch embedding function with recursive CUDA OOM retry.

looks_like_cuda_oom

recognition.looks_like_cuda_oom(exc)

Best-effort match for CUDA out-of-memory errors raised by ONNX Runtime.

try_compute_embeddings_batch

recognition.try_compute_embeddings_batch(
    compute_batch,
    bgr_aligned_batch,
    aligned_keypoints_batch=None,
)

Call a batch embedding function with recursive CUDA OOM retry.