Hi! I was wondering how images of different sizes are stacked together in a batch. I had a feeling that the content is copied to the top-left corner of a padded image tensor, and noticed that DataContainer is designed to deal with this. However DataContainer looks like a bare skeleton that doesn't have many operations defined, and I don't know where to find the collate function after Collect in the pipeline. How does the collate function work? And could you point me to the right place to check the implementation?
Hi! I was wondering how images of different sizes are stacked together in a batch. I had a feeling that the content is copied to the top-left corner of a padded image tensor, and noticed that
DataContainer
is designed to deal with this. HoweverDataContainer
looks like a bare skeleton that doesn't have many operations defined, and I don't know where to find the collate function afterCollect
in the pipeline. How does the collate function work? And could you point me to the right place to check the implementation?Thank you!