"Should transform.totensor be used, and when should it be used?"

[hz1z]

I saw someone's blog writing it this way, but I noticed that the official tutorial does not use transform.totensord. At first, I thought transform.totensord would convert my data from MetaTensor type to PyTorch's tensor type to save memory. However, I found that whether or not I use transform.totensord in the compose, the data ends up being MetaTensor type. Should I use transform.totensord? And when should it be used?"

[KumoLiu]

Hi @hz1z, thanks for your interest here.
If you want to convert your data to Tensor without meta information, you can do something like this:

ToTensord(track_meta=False)

Hope it helps, thanks.

[hz1z]

@KumoLiu I don’t want to convert metatensor to tensor.
Do you think it’s unnecessary to use totensor in his blog post?
I have two questions that I hope you can answer:

1. Do you recommend using transform.totensor in the compose?
2. If you would not use transform.totensor in the compose, then what is the purpose of this function? Thank you!!!

[KumoLiu]

Use it when you want to converts the input image to a tensor without applying any other transformations. If your dataset is already metatensor and you don't want to convert it to tensor, then no need to add this transform.
Thanks.