ResNet Backbone

tensoris.models.cv.backbone

Computer Vision ResNet Backbone Architecture.

Classes

VisionBackbone

Bases: Module

Convolutional Neural Network ResNet Vision Backbone Model.

References

He, K., Zhang, X., Ren, S., & Sun, J. (2016). Deep Residual Learning for Image Recognition. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2016), pp. 770-778. DOI: https://doi.org/10.1109/CVPR.2016.90 | arXiv: https://arxiv.org/abs/1512.03385

Source code in src/tensoris/models/cv/backbone.py

class VisionBackbone(nn.Module):
    """Convolutional Neural Network ResNet Vision Backbone Model.

    References:
        He, K., Zhang, X., Ren, S., & Sun, J. (2016).
        Deep Residual Learning for Image Recognition.
        Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2016), pp. 770-778.
        DOI: https://doi.org/10.1109/CVPR.2016.90 | arXiv: https://arxiv.org/abs/1512.03385
    """

    def __init__(
        self, in_channels: int = 3, num_classes: int = 1000, hidden_dim: int = 64
    ) -> None:
        """Initialize Vision Backbone model layers.

        Args:
            in_channels: Input image color channels (e.g. 3 for RGB).
            num_classes: Output class prediction logit dimension.
            hidden_dim: Convolutional channel dimension size.
        """
        super().__init__()
        self.stem = ConvStem(in_channels, hidden_dim)
        self.res_block1 = ResidualBlock(hidden_dim)
        self.res_block2 = ResidualBlock(hidden_dim)
        self.gap = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(hidden_dim, num_classes)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """Execute forward pass feature extraction.

        Args:
            x: Input image tensor batch of shape (N, C, H, W).

        Returns:
            Class prediction logits tensor of shape (N, num_classes).
        """
        features = self.stem(x)
        features = self.res_block1(features)
        features = self.res_block2(features)
        pooled = self.gap(features).flatten(1)
        return self.fc(pooled)

Methods:

__init__

__init__(in_channels=3, num_classes=1000, hidden_dim=64)

Initialize Vision Backbone model layers.

Parameters:

Name	Type	Description	Default
in_channels	int	Input image color channels (e.g. 3 for RGB).	3
num_classes	int	Output class prediction logit dimension.	1000
hidden_dim	int	Convolutional channel dimension size.	64

Source code in src/tensoris/models/cv/backbone.py

def __init__(
    self, in_channels: int = 3, num_classes: int = 1000, hidden_dim: int = 64
) -> None:
    """Initialize Vision Backbone model layers.

    Args:
        in_channels: Input image color channels (e.g. 3 for RGB).
        num_classes: Output class prediction logit dimension.
        hidden_dim: Convolutional channel dimension size.
    """
    super().__init__()
    self.stem = ConvStem(in_channels, hidden_dim)
    self.res_block1 = ResidualBlock(hidden_dim)
    self.res_block2 = ResidualBlock(hidden_dim)
    self.gap = nn.AdaptiveAvgPool2d((1, 1))
    self.fc = nn.Linear(hidden_dim, num_classes)

forward

forward(x)

Execute forward pass feature extraction.

Parameters:

Name	Type	Description	Default
x	Tensor	Input image tensor batch of shape (N, C, H, W).	required

Returns:

Type	Description
Tensor	Class prediction logits tensor of shape (N, num_classes).

Source code in src/tensoris/models/cv/backbone.py

def forward(self, x: torch.Tensor) -> torch.Tensor:
    """Execute forward pass feature extraction.

    Args:
        x: Input image tensor batch of shape (N, C, H, W).

    Returns:
        Class prediction logits tensor of shape (N, num_classes).
    """
    features = self.stem(x)
    features = self.res_block1(features)
    features = self.res_block2(features)
    pooled = self.gap(features).flatten(1)
    return self.fc(pooled)