Transforms

Eztorch supports transforms from Torchaug and from Torchvision.

On top of that, several video transforms have been defined.

Video transforms

• Video transforms
  • Video transforms
    • RemoveTimeDim
    • RandomResizedCrop
    • RandomTemporalDifference
    • TemporalDifference
  • SoccerNet
    • BatchReduceTimestamps
    • BatchMiddleTimestamps
    • ReduceTimestamps
    • MiddleTimestamps
  • Spot
    • BatchMiddleFrames
    • MiddleFrames
    • SpottingMixup

Apply key

class eztorch.transforms.ApplyTransformToKey(key, transform)

Applies transform to key of dictionary input.

Parameters:

• key (str) – The dictionary key the transform is applied to.

• transform (Module) – The transform that is applied.

class eztorch.transforms.ApplyTransformToKeyOnList(key, transform)

Applies transform to key of dictionary input where input is a list :type key: str :param key: the dictionary key the transform is applied to. :type transform: Module :param transform: the transform that is applied.

Example::

>>>  transforms.ApplyTransformToKeyOnList(
>>>       key='input',
>>>       transform=UniformTemporalSubsample(num_video_samples),
>>>   )

class eztorch.transforms.ApplySameTransformInputKeyOnList(transform, dim=1)

Apply same transform to the input list key.

Parameters:

• transform (Module) – The transform to apply.

• dim (int, optional) – The dimension to retrieve the various elements of the list.

  Default: 1

class eztorch.transforms.ApplySameTransformToKeyOnList(key, transform, dim=1)

Applies the same transform to key of dictionary input where input is a list.

Parameters:

• key (str) – the dictionary key the transform is applied to.

• transform (Module) – the transform that is applied.

• dim (int, optional) – The dimension to retrieve the various elements of the list.

  Default: 1

class eztorch.transforms.ApplyTransformAudioKey(transform)

Apply Transform to the audio key.

Parameters:

transform (Module) – The transform to apply.

class eztorch.transforms.ApplyTransformAudioKeyOnList(transform)

Apply Transform to the audio key.

Parameters:

transform (Module) – The transform to apply.

class eztorch.transforms.ApplyTransformInputKey(transform)

Apply Transform to the input key.

Parameters:

transform (Module) – The transform to apply.

class eztorch.transforms.ApplyTransformInputKeyOnList(transform)

Apply Transform to the input key.

Parameters:

transform (Module) – The transform to apply.

class eztorch.transforms.ApplyTransformOnDict(transform)

Apply Transform to the audio key.

Parameters:

transform (Module) – The transform to apply.

Apply transform on list

class eztorch.transforms.ApplyTransformOnList(transform, list_len=2)

Apply transform to a list of input.

Parameters:

• transform (Module) – A transform for the list of input.

• list_len (int, optional) – len of the input.

  Default: 2

class eztorch.transforms.ApplyTransformsOnList(transforms)

Apply transform to a list of input.

Parameters:

• transform – A transform for the list of input.

• list_len – len of the input.

Dict keep keys

class eztorch.transforms.DictKeepKeys(keys)

Keep specified keys in dict.

References

• https://github.com/kalyanvasudev/pytorchInput-1/blob/export-D33431232/pytorchInput_trainer/pytorchInput_trainer/datamodule/transforms.py

Parameters:

keys (List[str]) – The list of keys to keep.

class eztorch.transforms.DictKeepInputLabel

Transform dict to list containing values of only 'input' and 'label'.

class eztorch.transforms.DictKeepInputLabelIdx

Transform dict to list containing values of only 'input', 'label', 'index' and 'aug_index'.

Div 255

class eztorch.transforms.Div255Input(inplace=True, dtype=torch.float32)

Perform Division by 255 on a tensor or list of tensor.

Multi-crop

class eztorch.transforms.MultiCropTransform(set_transforms)

Define multi crop transform that apply several sets of transform to the inputs.

Parameters:

set_transforms (List[Any]) – List of Dictionary of sets of transforms specifying transforms and number of views per set.

Example:

set_transforms = [
    {'transform': [...], 'num_views': ...},
    {'transform': [...], 'num_views': ...},
    ...
]

transform = MultiCropTransform(
    set_transforms
)

Only input transform

class eztorch.transforms.OnlyInputTransform(transform)

Apply Transform to only the key 'input' in a sample dictionary.

Parameters:

transform (Callable) – The transform to apply.

class eztorch.transforms.OnlyInputListTransform(transform)

Apply Transform to only the key 'input' in a list of sample dictionary.

Parameters:

transform (Callable) – The transform to apply.

class eztorch.transforms.OnlyInputListSameTransform(transform)

Apply the same transform to only the key 'input' in a list of sample dictionary.

Parameters:

transform (Callable) – The transform to apply.

class eztorch.transforms.OnlyInputTransformWithDictTransform(transform, dict_transform, first_dict=False)

Apply Transform to only the key 'input' in a sample dictionary with a transformation on the dictionary afterwards.

Parameters:

• transform (Callable) – The transform to apply to the input.

• dict_transform (Callable) – The transform to apply to the dictionary.

• first_dict (bool, optional) – If True, first apply the transformation on the dict, else, first apply the transformation on the input.

  Default: False

class eztorch.transforms.OnlyInputListTransformWithDictTransform(transform, dict_transform, first_dict=False)

Apply Transform to only the key 'input' in a list of sample dictionary with a transformation on the dictionary afterwards.

Parameters:

• transform (Callable) – The transform to apply to the input.

• dict_transform (Callable) – The transform to apply to the dictionary.

• first_dict (bool, optional) – If True, first apply the transformation on the dict, else, first apply the transformation on the input.

  Default: False

Random Resized Crop

class eztorch.transforms.RandomResizedCrop(size, scale=[0.08, 1.0], ratio=[0.75, 1.3333333333333333], interpolation='bilinear', antialias=True, **kwargs)

Remove key

class eztorch.transforms.RemoveKey(key)

Removes the given key from the input dict. Useful for removing modalities from a video clip that aren’t needed.

Parameters:

key (str) – The dictionary key to remove.

class eztorch.transforms.RemoveAudioKey

Remove audio key from sample dictionary.

class eztorch.transforms.RemoveInputKey

Remove video key from sample dictionary.