How to change the output size of the model when the batch size is specified?

vision
1

Hi I’m trying to calculate distance between embedding vectors which is extracted from the model. I’m working on image retrieval problem.
when I work with keras Imagedatagenetor function, even when the batch size is specified, the output shape of the model is for example (number of query images or reference images, num_classes)

but when I work with pytorch, when using dataloader with batchsize say 32, the output size of the model becomes (32, num_classes). The last layer is linear layer.

I need shape of (number of query images or reference images, num_classes) to do the samething as I did with keras because I need embedding vectors each to compute the distance.
Any advice please? thanks

2

Have you tried something like this?

outputs = []
for input_batch in dataloader:
      output_batch = model(input_batch) #torch Tensor of shape (32, num_classes)
      outputs.append(output_batch)
print(outputs.shape) #Torch Tensor of shape (number of query images, num_classes)
3

isn’t the output of the model the shape of tensor? error says that there are no attribute ‘append’. how can I solve this??

4

I simply created a list and appended each output batch to my list. Can you share your code here?