Dev Tools · 1h ago
Testing OpenCV SFace for Face Embedding Nearest Neighbor Search
A developer tested OpenCV Zoo's SFace model to generate 128-dimensional face embeddings and verify nearest neighbor search accuracy. Using Olivetti Faces and LFW datasets, the test checked whether same-person images become the 1-nearest-neighbor and compared cosine similarity between same and different person pairs. The model, licensed under Apache-2.0, outputs embeddings from 112x112 aligned BGR face crops.
Meridian48 take
The post is a practical verification of a well-known model, but lacks novel findings or performance benchmarks against other methods.
Read the full reporting
Testing Face Embedding Nearest Neighbor Search with OpenCV SFace →
DEV Community
face-recognitionopencv