search

arsfutura / face-recognition

A framework for creating and using a Face Recognition system.
BSD 3-Clause "New" or "Revised" License
146 stars 49 forks source link

Incorrect classifcation #17

Closed vinayak-parab closed 3 years ago

vinayak-parab commented 4 years ago

I tried to put 5 different identities and generated their embedding and then trained the model using those embedding. However, it is unable to differentiate between multiple classes. When I try to run the classifier.py and provide an image path it incorrectly detecting all the images with my name instead of other identity. Can you please suggest if I am missing something? or the model got over fitted to the data?

ldulcic commented 4 years ago

Hi @vinayak-parab! If I understand this correctly, you used only 5 images to train the model? Unfortunately that's too few, you need at least couple of images per identity.

vinayak-parab commented 4 years ago

@ldulcic No. I used 5 images of each identity and I had 5 such different identities.

ldulcic commented 4 years ago

@vinayak-parab can you post the output of training?

prakharter commented 3 years ago

Hi , i have observed that creating the embeddings first before training and using those created embeddings in training results in wrong inference , however using the same dataset to train directly results in correct inference ?, Also is there a way to add to existing embeddings rather than creating new ones every time with the entire dataset ? Thanks in advance

ldulcic commented 3 years ago

Hi @prakharter! Could you post here the exact scripts you called (with parameters used) when you got wrong inference?

ldulcic commented 3 years ago

Closing due to inactivity.