thomaspillot/kraken_rec

KrakenID Entrance Recognition

Internal smoke-test artifact for KrakenID entrance-camera recognition experiments.

This repository contains a compact recognition checkpoint, preprocessing metadata, and a small validation image set used by the entrance-controller integration test suite.

Repository Layout

• config.json: model architecture and classifier metadata.
• preprocessor_config.json: image normalization and resize settings.
• threshold.json: integration threshold used by smoke tests.
• labels.json: class labels returned by the integration wrapper.
• model.safetensors: compact checkpoint artifact for loader compatibility.
• test_examples/accepted/: positive smoke-test examples.
• test_examples/rejected/: negative smoke-test examples.
• eval/: summary metrics from the latest validation batch.
• scripts/run_smoke_test.py: local validation helper.

Test Examples

test_examples/accepted/ contains images expected to pass the current entrance recognition threshold.

test_examples/rejected/ contains negative controls expected to fail.

The examples are intended for integration testing only and are not a production identity roster.

Intended Use

The entrance controller downloads this repository as a model bundle during integration testing. The live controller applies its own authorization policy after recognition and should treat this repository as a model artifact, not as a source of operational secrets.

Version

• Model family: krakenid-entrance-recognition
• Export profile: edge-smoke
• Input size: 224x224
• Distance metric: cosine
• Last validation target: entrance-camera still images