simet.transforms.sample_downstream

simet.transforms.sample_downstream

SampleDownstreamTransform

Bases: Transform

Lightweight preprocessing for downstream tasks (64×64, [-1, 1]).

Produces a torchvision.transforms.Compose that: 1) resizes the shorter side to 64 px, 2) center-crops to 64×64, 3) converts to a tensor, 4) normalizes channels to approximately [-1, 1] using mean=0.5/std=0.5.

Use this for downstream or toy models that operate at 64×64, not for Inception-v3 feature extraction (which expects 299×299 and typically ImageNet mean/std).

Example

tfm = SampleDownstreamTransform().get_transform() x = tfm(PIL_image) # tensor shape [3, 64, 64], roughly in [-1, 1]

get_transform

get_transform()

Return the 64×64 downstream preprocessing pipeline.

Returns:

Name	Type	Description
Compose	Compose	Resize→CenterCrop→ToTensor→Normalize(mean=0.5, std=0.5).

Source code in simet/transforms/sample_downstream.py

@override
def get_transform(self) -> Compose:
    """Return the 64×64 downstream preprocessing pipeline.

    Returns:
        Compose: Resize→CenterCrop→ToTensor→Normalize(mean=0.5, std=0.5).
    """
    return Compose(
        [
            transforms.Resize(64),        # 64×64 target size for small models
            transforms.CenterCrop(64),    # Ensures centered 64×64 crop
            transforms.ToTensor(),        # PIL → Tensor in [0, 1]
            transforms.Normalize(
                mean=[0.5, 0.5, 0.5],
                std=[0.5, 0.5, 0.5],
            ),  # maps to approximately [-1, 1]
        ]
    )