VGG_Face works better. If I compare a photo of the same person with and without glasses, then VHG gives the best result. Other models generally say that they are not even alike.
if your question is for deepface, then no, it offers just pre-trained weights on the other hand, you can find how to train videos and tutorials in my blog
Right on! Thank you for all your work! I'm going to throw you a sub on Paetron this week. Such a cool tool. Currently I've been using Facenet with mediapipe. I'm guessing mediapipe is missing detection on a lot of faces? @@sefiks
Also if getting images from all over the place. Thousands taken with random coloring, lighting, angles, etc. Should I be using align=True to generate a representation of them all?@@sefiks
do i need to have nvidia to use deepface? because i havn't nvidia and gpu in my linux server and i get errors: - Could not load dynamic library 'libcudart.so.11.0' - Could not load dynamic library 'libnvinfer.so.7' - Could not load dynamic library 'libnvinfer_plugin.so.7 - Warning: Cannot dlopen some TensorRT libraries.
I don't know how they calculated the accuracy of these models, but VGGface is the best.
VGG_Face works better. If I compare a photo of the same person with and without glasses, then VHG gives the best result. Other models generally say that they are not even alike.
Thnak brother i try many Face Recognition library but I get best result from deepface thank you.
Then, please star the repo to support the project. Thank you.
Very helpful video, thank you!!
This video is really helpful. Thank you Sefik !
You're welcome!
do you have videos on how to train, retrain with frozen layers one of these models?
if your question is for deepface, then no, it offers just pre-trained weights
on the other hand, you can find how to train videos and tutorials in my blog
Which combination of model and detector_backend do you feel works the best? In terms of accuracy and speed?
Facenet + mtcnn
Right on! Thank you for all your work! I'm going to throw you a sub on Paetron this week. Such a cool tool.
Currently I've been using Facenet with mediapipe. I'm guessing mediapipe is missing detection on a lot of faces?
@@sefiks
Also if getting images from all over the place. Thousands taken with random coloring, lighting, angles, etc. Should I be using align=True to generate a representation of them all?@@sefiks
@@SS-gu2tx alignment increases accuracy
Very nice video. I have a doubt: is the OpenFace's structure the same as FaceNet's structure? What would be the diffence between them?
Structure is different, the data set they were trained different as well.
@@sefiks ok but why did you put in the video OpenFace and FaceNet with the same structure?
They are both based on resnet but have little differences
What about arcface?
ArcFace shows a close score to FaceNet. I haven't implement it when I record this video. That's why, it is not mentioned here.
@@sefiks Would you know what are the benefits of ArcFace in comparison with FaceNet? FaceNet is considered an old algorithm.
@@tanpengshi The both FaceNet and ArcFace have more than 99% accuracy. The new is not always better :)
what about vggface2
vggface2 is a facial data set. it is not a face recognition model. you can train any face recognition model with vggface2.
@@sefiks Ohk... That clears my doubt. I was wondering why VGGFace2 isn't covered over here.
That helps ! Thank you !
do i need to have nvidia to use deepface? because i havn't nvidia and gpu in my linux server and i get errors:
- Could not load dynamic library 'libcudart.so.11.0'
- Could not load dynamic library 'libnvinfer.so.7'
- Could not load dynamic library 'libnvinfer_plugin.so.7
- Warning: Cannot dlopen some TensorRT libraries.
No, gpu is not a must. Your tensorflow installation seems problematic.