UniFace: A Unified Face Analysis Library for Python | Detection, alignment, landmarks, face-mesh, recognition, parsing, gaze, attributes and anti-spoofing under one API.
Python computer-visiondeep-learningface-alignmentface-analysisface-anonymizationface-anti-spoofingface-attributesface-detectionface-meshface-parsingface-quality-assessmentface-recognitionface-segmentationface-trackingfacial-landmarksgaze-estimationhead-pose-estimationonnxruntimeportrait-mattingpython
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Face Vector Store with FAISS\n",
"\n",
"[](https://colab.research.google.com/github/yakhyo/uniface/blob/main/examples/10_face_vector_store.ipynb) [](https://www.kaggle.com/code/yakhyokhuja/face-vector-store-with-faiss)\n",
"\n",
"Enrol faces into a FAISS index, search it, and save it to disk.\n",
"Unlike direct pairwise comparison (see `04_face_search`), a vector store lets you efficiently index\n",
"thousands of face embeddings and retrieve the closest match in sub-millisecond time.\n",
"\n",
"---\n",
"\n",
"**UniFace** is a lightweight, production-ready Python library for face detection, recognition, tracking, landmark analysis, face parsing, gaze estimation, and face attributes.\n",
"\n",
"GitHub: [github.com/yakhyo/uniface](https://github.com/yakhyo/uniface) | Docs: [yakhyo.github.io/uniface](https://yakhyo.github.io/uniface)\n",
"\n",
"## Setup\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"%pip install -q \"uniface[cpu]\" faiss-cpu\n",
"\n",
"# Clone repo for assets (Colab / Kaggle only)\n",
"import os\n",
"if any(k in os.environ for k in ('COLAB_GPU', 'COLAB_RELEASE_TAG', 'KAGGLE_KERNEL_RUN_TYPE')):\n",
" if not os.path.exists('uniface'):\n",
" !git clone --depth 1 https://github.com/yakhyo/uniface.git\n",
" os.chdir('uniface/examples')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import cv2\n",
"import matplotlib.pyplot as plt\n",
"import shutil\n",
"\n",
"import uniface\n",
"from uniface.analyzer import FaceAnalyzer\n",
"from uniface.detection import RetinaFace\n",
"from uniface.recognition import ArcFace\n",
"from uniface.stores import FAISS\n",
"\n",
"print(f'UniFace version: {uniface.__version__}')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"analyzer = FaceAnalyzer(\n",
" detector=RetinaFace(confidence_threshold=0.5),\n",
" recognizer=ArcFace(),\n",
")\n",
"\n",
"DB_PATH = './demo_face_index'\n",
"store = FAISS(embedding_size=512, db_path=DB_PATH)\n",
"print(store)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 1. Enroll Faces into the Store\n",
"\n",
"We detect faces in the test images and add each embedding with metadata."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"enrollment_images = {\n",
" '../assets/source/verify_now_2010.jpg': 'person_a',\n",
" '../assets/source/age_child.jpg': 'person_b',\n",
" '../assets/source/age_adult.jpg': 'person_c',\n",
" '../assets/source/age_middle.jpg': 'person_d',\n",
" '../assets/source/age_senior.jpg': 'person_e',\n",
"}\n",
"\n",
"for path, label in enrollment_images.items():\n",
" image = cv2.imread(path)\n",
" faces = analyzer.analyze(image)\n",
" if faces:\n",
" store.add(\n",
" embedding=faces[0].embedding,\n",
" metadata={'label': label, 'source': path},\n",
" )\n",
" print(f'Enrolled {label} from {path}')\n",
"\n",
"print(f'\\nStore size: {store.size} vectors')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 2. Search the Store\n",
"\n",
"Use a query image to find the closest match in the database."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"query_image = cv2.imread('../assets/source/verify_now_2024.jpg')\n",
"query_faces = analyzer.analyze(query_image)\n",
"\n",
"if query_faces:\n",
" result, similarity = store.search(query_faces[0].embedding, threshold=0.4)\n",
"\n",
" if result:\n",
" print(f'Match found: {result[\"label\"]} (similarity: {similarity:.4f})')\n",
" print(f'Source: {result[\"source\"]}')\n",
" else:\n",
" print(f'No match above threshold (best similarity: {similarity:.4f})')"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"if query_faces and result:\n",
" matched_image = cv2.imread(result['source'])\n",
"\n",
" fig, axes = plt.subplots(1, 2, figsize=(10, 4))\n",
" axes[0].imshow(cv2.cvtColor(query_image, cv2.COLOR_BGR2RGB))\n",
" axes[0].set_title('Query', fontsize=12)\n",
" axes[1].imshow(cv2.cvtColor(matched_image, cv2.COLOR_BGR2RGB))\n",
" axes[1].set_title(f'Match: {result[\"label\"]} ({similarity:.3f})', fontsize=12)\n",
" for ax in axes:\n",
" ax.axis('off')\n",
" plt.tight_layout()\n",
" plt.show()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 3. Save and Reload the Index\n",
"\n",
"The index and metadata can be persisted to disk and loaded later."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"store.save()\n",
"\n",
"# Create a fresh store and load the saved data\n",
"store_reloaded = FAISS(embedding_size=512, db_path=DB_PATH)\n",
"loaded = store_reloaded.load()\n",
"print(f'Load successful: {loaded}')\n",
"print(f'Reloaded store size: {store_reloaded.size} vectors')\n",
"\n",
"# Verify search still works after reload\n",
"if query_faces:\n",
" result, similarity = store_reloaded.search(query_faces[0].embedding, threshold=0.4)\n",
" if result:\n",
" print(f'Search after reload: {result[\"label\"]} ({similarity:.4f})')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 4. Remove Entries\n",
"\n",
"Remove all entries matching a metadata key-value pair."
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"print(f'Before removal: {store.size} vectors')\n",
"\n",
"removed = store.remove(key='label', value='person_0')\n",
"print(f'Removed {removed} entry')\n",
"print(f'After removal: {store.size} vectors')\n",
"\n",
"# Searching for the removed person should now return a different (lower) match\n",
"if query_faces:\n",
" result, similarity = store.search(query_faces[0].embedding, threshold=0.4)\n",
" if result:\n",
" print(f'\\nClosest remaining match: {result[\"label\"]} ({similarity:.4f})')\n",
" else:\n",
" print(f'\\nNo match above threshold (best similarity: {similarity:.4f})')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 5. Cleanup"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"shutil.rmtree(DB_PATH, ignore_errors=True)\n",
"print('Cleaned up demo index.')"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Notes\n",
"\n",
"- Embeddings must be L2-normalised before you add them. ArcFace already returns normalised ones.\n",
"- The index is `IndexFlatIP`, so an inner product equals the cosine similarity.\n",
"- `search()` returns only the single best match above the threshold, or `None`."
]
}
],
"metadata": {
"kernelspec": {
"display_name": "base",
"language": "python",
"name": "python3"
},
"language_info": {
"codemirror_mode": {
"name": "ipython",
"version": 3
},
"file_extension": ".py",
"mimetype": "text/x-python",
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.13.5"
}
},
"nbformat": 4,
"nbformat_minor": 4
}