Skip to content

spencerwooo/torch-featurelayer

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

16 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Ruff image image image lint test

torch-featurelayer

🧠 Simple utility functions and wrappers for hooking onto layers within PyTorch models for feature extraction.

Tip

This library is intended to be a simplified and well-documented implementation for extracting a PyTorch model's intermediate layer output(s). For a more sophisticated and complete implementation, either consider using torchvision.models.feature_extraction, or check the official torch.fx.

Install

pip install torch-featurelayer

Usage

Imports:

import torch
from torchvision.models import vgg11
from torch_featurelayer import FeatureLayer

Load a pretrained VGG-11 model:

model = vgg11(weights='DEFAULT').eval()

Hook onto layer features.15 of the model:

layer_path = 'features.15'
hooked_model = FeatureLayer(model, layer_path)

Forward pass an input tensor through the model:

x = torch.randn(1, 3, 224, 224)
feature_output, output = hooked_model(x)

feature_output is the output of layer features.15. Print the output shape:

print(f'Feature layer output shape: {feature_output.shape}')  # [1, 512, 14, 14]
print(f'Model output shape: {output.shape}')  # [1, 1000]

Check the examples directory for more.

API

torch_featurelayer.FeatureLayer

The FeatureLayer class wraps a model and provides a hook to access the output of a specific feature layer.

  • __init__(self, model: torch.nn.Module, feature_layer_path: str)

    Initializes the FeatureLayer instance.

    • model: The model containing the feature layer.
    • feature_layer_path: The path to the feature layer in the model.
  • __call__(self, *args: Any, **kwargs: Any) -> tuple[torch.Tensor | None, torch.Tensor]

    Performs a forward pass through the model and updates the hooked feature layer.

    • *args: Variable length argument list.
    • **kwargs: Arbitrary keyword arguments.

    Returns a tuple containing the feature layer output and the model output.

torch_featurelayer.FeatureLayers

The FeatureLayers class wraps a model and provides hooks to access the output of multiple feature layers.

  • __init__(self, model: torch.nn.Module, feature_layer_paths: list[str])

    Initializes the FeatureLayers instance.

    • model: The model containing the feature layers.
    • feature_layer_paths: A list of paths to the feature layers in the model.
  • __call__(self, *args: Any, **kwargs: Any) -> tuple[dict[str, torch.Tensor | None], torch.Tensor]

    Performs a forward pass through the model and updates the hooked feature layers.

    • *args: Variable length argument list.
    • **kwargs: Arbitrary keyword arguments.

    Returns a tuple containing the feature layer outputs and the model output.

torch_featurelayer.get_layer_candidates(module: torch.nn.Module, max_depth: int = 1) -> Generator[str, None, None]

The get_layer_candidates function returns a generator of layer paths for a given model up to a specified depth.

  • model: The model to get layer paths from.
  • max_depth: The maximum depth to traverse in the model's layers.

Returns a generator of layer paths.

License

MIT