Khalyie/lfw-face-recognition-data
LFW Face Recognition — Aligned Faces, Embeddings & Test Predictions Companion data for Khalyie/lfw-face-recognition. Files aligned_faces.pt — MTCNN-aligned 160x160 uint8 face tensors for all 5749+ identities / ~13k images, indexed to match metadata.json's identity/label arrays alignment_fallback_indices.json — indices where MTCNN failed to detect a face and a center-crop fallback was used metadata.json — identity list + train/val/test index splits… See the full description on the dataset page: https://huggingface.co/datasets/Khalyie/lfw-face-recognition-data.
LFW Face Recognition — Aligned Faces, Embeddings & Test Predictions
Companion data for `Khalyie/lfw-face-recognition`.
Files
aligned_faces.pt— MTCNN-aligned 160x160 uint8 face tensors for all 5749+ identities / ~13k images, indexed to matchmetadata.json's identity/label arraysalignment_fallback_indices.json— indices where MTCNN failed to detect a face and a center-crop fallback was usedmetadata.json— identity list + train/val/test index splitstrain/val/test_embeddings.npy,*_labels.npy,*_indices.npy— 512-dim L2-normalized embeddings per split, with original dataset indices preservedtest_predictions.csv— per-image predictions from the test set: true identity, predicted identity, similarity score, correct/incorrect, top-3 candidatestest_predictions_sample.png— visual grid of sample predictions
Reload embeddings/labels with numpy.load(...), and metadata.json's train_indices / val_indices / test_indices to know which original LFW image each row corresponds to.
