How to do inference using grid sampler and how does it exactly work?

Hi All,

I want to perform inference on a multi modal data using the grid sampler. However, I am not able to wrap my head around the working of the grid sampler in this link: https://torchio.readthedocs.io/data/patch_inference.html?highlight=gridsample

Here, it says that:

grid_sampler = torchio.inference.GridSampler(
    sample,
    patch_size,
    patch_overlap,
)

Here the sample is one individual object of the Subject class? I was just wondering if this would become a little clumsy while performing inference on a list of subjects?Or do I just create a list of grid_samplers and iterate through them?

It would be great if you could point me to an example which does inference for an entire list of images.

Thanks you!

[meghbhalerao]

Hi @fepegar, Thanks a lot for your explanation. I understood it.

[fepegar]

What I would do is something like:

for subject in dataset:
    subject_grid_sampler = GridSampler(subject, ...)  # the patches positions are computed here
    aggregator = GridAggregator(subject_grid_sampler)  # the aggregator checks the sampler to take decisions about overlap or padding
    loader = DataLoader(subject_grid_sampler, batch_size=B)
    for batch in loader:
        inputs, locations = prepare_batch(batch)
        outputs = model(input)    
        aggregator.add_batch(outputs, locations)
    output_tensor = aggregator.get_output_tensor()

Why would you need a data loader for the subjects? Would you like to load them in parallel? In that case, you can use a DataLoader with collate_fn=lambda x: x and the subjects data loader will return a list of instances of Subject that you can iterate through, using another data loader to run the inference on multiple patches at the same time.