search

serengil / deepface

A Lightweight Face Recognition and Facial Attribute Analysis (Age, Gender, Emotion and Race) Library for Python
https://bit.ly/deepface-py
MIT License
14.71k stars 2.23k forks source link

Request for documentation on Face Recognition models architectures and training datasets #1367

Closed HectorPenades closed 3 weeks ago

HectorPenades commented 1 month ago

Description

Dear DeepFace Team,

I hope this message finds you well. I would like to kindly request documentation on the specific models and weights used in DeepFace for face recognition. Could you provide details on the exact architectures (e.g. VGG, Inception, DNN, Resnet34, ResNet 50, ResNet100, ...) and the training datasets (e.g., MS1M-Celeb, VGGFace2, CASIA-WebFace, etc.) used for each model integrated into DeepFace? This would greatly help the community understand the underlying implementations and facilitate more accurate comparisons in research.

Thank you for your consideration, and I look forward to your response.

HectorPenades commented 1 month ago

I will add in this comment the information that I am finding:

CONFIRMED

NOT CONFIRMED ArcFace: ResNet34, CASIA-WebFace, 40 E - Link: https://github.com/leondgarse/Keras_insightface FaceNet: Inception, CASIA-WebFace or VGGFace2, X E - Link: https://github.com/davidsandberg/facenet VGG-Face: VGG-16, VGG-Face

PENDING Facenet512: Inception OpenFace: DeepFace: DNN, Facebook Dataset DeepID: DNN, CelebFaces+ / WDRef ArcFace: ResNet34, CASIA-WebFace Dlib: SFace: GhostFaceNet: GhostNet

serengil commented 3 weeks ago
Model Backbone Source URL
VGG-Face VGG-Face https://www.robots.ox.ac.uk/~vgg/software/vgg_face/src/vgg_face_matconvnet.tar.gz
FaceNet128d Inception ResNet v1 https://drive.google.com/file/d/0B5MzpY9kBtDVZ2RpVDYwWmxoSUk
FaceNet512d Inception ResNet v1 https://drive.google.com/open?id=1EXPBSXwTaqrSC0OhUdXNmKSh9qJUQ55-
ArcFace ResNet34 https://github.com/leondgarse/Keras_insightface/releases/download/v1.0.0/resnet34_MXNET_E_SGD_REG_1e3_on_batch_true_lr1e1_random0_arc_S32_E1_BS512_casia_basic_agedb_30_epoch_36_0.949500.h5
OpenFace Inception Resnet V1 https://github.com/vsyw/Keras-OpenFace/blob/master/model/nn4.small2.v1.h5
DeepId DeepID2 https://github.com/ruoyiran/DeepID/blob/master/model/best_model.ckpt.data-00000-of-00001
Dlib ResNetV1 http://dlib.net/files/dlib_face_recognition_resnet_model_v1.dat.bz2
SFace MobileFaceNet https://github.com/opencv/opencv_zoo/raw/main/models/face_recognition_sface/face_recognition_sface_2021dec.onnx
GhostFaceNet GhostFaceNetV1-1.3-1 (A) https://github.com/HamadYA/GhostFaceNets/releases/download/v1.2/GhostFaceNet_W1.3_S1_ArcFace.h5
HectorPenades commented 6 days ago

Thank you very much for providing details on the architectures used in DeepFace. Your prompt response is greatly appreciated.

If possible, could you also share information on the specific datasets used to train each model integrated into DeepFace? Knowing more about the datasets, such as whether they include MS1M-Celeb, VGGFace2, CASIA-WebFace, or others, would be highly valuable for the research community. This would enhance our understanding of the model performances and help improve comparisons in related studies.

Once again, thank you for your assistance, and I look forward to your reply.

@serengil