facenet-pytorch
by timesformer
Deep face recognition with triplet loss embeddings
timesformer/facenet-pytorchmixpeek://face_identity@v1/timesformer_facenet_v1Overview
FaceNet maps face images to a compact 128-dimensional embedding space where distances directly correspond to face similarity. Trained using triplet loss, it achieves 99.63% accuracy on the Labeled Faces in the Wild benchmark.
On Mixpeek, FaceNet provides face embedding extraction for identity-based search, find all appearances of a person across your video and image library.
Architecture
InceptionResnetV1 backbone fine-tuned with triplet loss on VGGFace2 dataset. Produces 512-dim or 128-dim face embeddings. Pre-processes with MTCNN face detection and alignment.
Mixpeek SDK Integration
import { Mixpeek } from "mixpeek";const mx = new Mixpeek({ apiKey: "API_KEY" });await mx.collections.ingest({collection_id: "my-collection",source: { url: "https://example.com/video.mp4" },feature_extractors: [{name: "face_detection",version: "v1",params: {model_id: "timesformer/facenet-pytorch"}}]});
Capabilities
- Face verification (same/different person)
- Face identification across large galleries
- 128-dim or 512-dim face embeddings
- Built-in MTCNN face alignment
Use Cases on Mixpeek
Benchmarks
| Dataset | Metric | Score | Source |
|---|---|---|---|
| LFW | Accuracy | 99.63% | Schroff et al., 2015 — Table 4 |
| CFP-FP | Accuracy | 98.12% | FaceNet PyTorch model card |
Performance
Specification
Research Paper
FaceNet: A Unified Embedding for Face Recognition and Clustering
arxiv.orgBuild a pipeline with facenet-pytorch
Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.
Open Pipeline Builder