How i can save feature vectors for each image in my folder

1

hi friends
I trained my model in the following way:
Now i want to seed my training data into this trained model again and extract the feature vector for each image so that its size is “embed-size” in my code. And then I want to save all these feature vectors in my hard disk .On the other hand, I have a lot of image folders and I want to save the image vectors in the way similar to those image folders and with the same names.

Pleas help me how to accomplish this via code?

9

2

Hey @romina.ui,

Why not just save the output of f_anc, f_pos, f_neg as numpy array using np.save()?

3

I don’t need f_pos and f_neg after train my model. I want just f_nac for save my image. should i change my code?
for training stage and I select my training data as f_anc , f_pos and f_neg but for extract my feature vector and save them i want seed all images again to my trained model and save feature vectors in new folders that they have same names with my image folder

4

After training the model, you can redefine the forward function to return only f_nac. You can use “types” library to redefine the member functions of an existing object. Is this what you are looking for?

5

Yes .but I think my idea is not true . so I think I should load trained model without any change .
I write this code for load trained model with its weight but its not true.
Can you help me Please?

model.load_state_dict(torch.load(“trained model path” + “best weights path”))