faceverify

Performance Tuning

Optimize FaceVerify for your use case.

---

Benchmarks

Approximate performance on CPU (Intel i7):

Backend	Detection Time	Notes
OpenCV	~20ms	Fastest
MediaPipe	~50ms	Good balance
MTCNN	~100ms	Accurate
RetinaFace	~200ms	Most accurate

Model	Embedding Time	Notes
Facenet	~150ms	Default
ArcFace	~180ms	Similar
VGGFace	~250ms	Larger model

---

Optimization Strategies

1. Choose Appropriate Backend

For real-time applications:

config = VerifierConfig(detector_backend="opencv")

For batch processing:

config = VerifierConfig(detector_backend="retinaface")

2. Enable GPU

config = VerifierConfig(enable_gpu=True)

[!IMPORTANT] Requires CUDA-compatible GPU and TensorFlow with GPU support.

3. Batch Processing

Process multiple pairs at once:

pairs = [("a1.jpg", "a2.jpg"), ("b1.jpg", "b2.jpg"), ...]
results = verifier.verify_batch(pairs, parallel=True)

4. Cache Embeddings

Extract and store embeddings for repeated comparisons:

# Extract once
embeddings = {}
for image_path in images:
    result = verifier.extract_embedding(image_path)
    embeddings[image_path] = result.embedding

# Compare many times
similarity, _ = verifier.compare_embeddings(
    embeddings["person1.jpg"],
    embeddings["person2.jpg"]
)

5. Reduce Image Size

Resize large images before processing:

import cv2

image = cv2.imread("large_image.jpg")
resized = cv2.resize(image, (640, 480))
result = verifier.verify(resized, "other.jpg")

---

Memory Optimization

Limit Batch Size

config = VerifierConfig(batch_size=8)  # Reduce if memory limited

Process Sequentially

results = verifier.verify_batch(pairs, parallel=False)

---

Profiling

Measure processing time:

```python import time

start = time.time() result = verifier.verify(“a.jpg”, “b.jpg”) elapsed = time.time() - start

print(f”Verification took: {elapsed:.3f}s”) print(f”Internal time: {result.processing_time:.3f}s”)

This site is open source. Improve this page.