Source code for mmaction.models.heads.tsn_audio_head

# Copyright (c) OpenMMLab. All rights reserved.
import torch
import torch.nn as nn
from mmengine.model.weight_init import normal_init

from mmaction.registry import MODELS
from mmaction.utils import ConfigType
from .base import BaseHead


@MODELS.register_module()
class TSNAudioHead(BaseHead):
    """Classification head for TSN on audio.

    Args:
        num_classes (int): Number of classes to be classified.
        in_channels (int): Number of channels in input feature.
        loss_cls (Union[dict, ConfigDict]): Config for building loss.
            Defaults to ``dict(type='CrossEntropyLoss')``.
        spatial_type (str): Pooling type in spatial dimension.
            Defaults to ``avg``.
        dropout_ratio (float): Probability of dropout layer. Defaults to 0.4.
        init_std (float): Std value for Initiation. Defaults to 0.01.
    """

    def __init__(self,
                 num_classes: int,
                 in_channels: int,
                 loss_cls: ConfigType = dict(type='CrossEntropyLoss'),
                 spatial_type: str = 'avg',
                 dropout_ratio: float = 0.4,
                 init_std: float = 0.01,
                 **kwargs) -> None:
        super().__init__(num_classes, in_channels, loss_cls=loss_cls, **kwargs)

        self.spatial_type = spatial_type
        self.dropout_ratio = dropout_ratio
        self.init_std = init_std

        if self.spatial_type == 'avg':
            # use `nn.AdaptiveAvgPool2d` to adaptively match the in_channels.
            self.avg_pool = nn.AdaptiveAvgPool2d((1, 1))
        else:
            self.avg_pool = None

        if self.dropout_ratio != 0:
            self.dropout = nn.Dropout(p=self.dropout_ratio)
        else:
            self.dropout = None
        self.fc_cls = nn.Linear(self.in_channels, self.num_classes)

    def init_weights(self) -> None:
        """Initiate the parameters from scratch."""
        normal_init(self.fc_cls, std=self.init_std)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        """Defines the computation performed at every call.

        Args:
            x (torch.Tensor): The input data.

        Returns:
            torch.Tensor: The classification scores for input samples.
        """
        # [N * num_segs, in_channels, h, w]
        x = self.avg_pool(x)
        # [N, in_channels, 1, 1]
        x = x.view(x.size(0), -1)
        # [N, in_channels]
        if self.dropout is not None:
            x = self.dropout(x)
        # [N, in_channels]
        cls_score = self.fc_cls(x)
        # [N, num_classes]
        return cls_score