ClassificationResNet

class ClassificationResNet(

in_channels,

out_channels,

num_blocks,

hidden_dim,

data_dim,

in_proj_cfg,

out_proj_cfg,

norm_cfg,

block_cfg,

dropout_in_cfg,

condition_in_proj_cfg=None,

target_size=None,

gradient_checkpointing=False,

)

Bases: ResidualNetwork

Residual network with global-average-pool readout for classification.

Inherits the full constructor and backbone from ResidualNetwork. Overrides only forward() to replace the spatial output with a single class-logit vector via global average pooling.

Output shape: [B, out_channels] regardless of input spatial size — the model is therefore resolution-agnostic at inference time.

No ``target_size``: the inherited target_size attribute is ignored; GAP serves as the spatial aggregation step.

All constructor arguments are documented in ResidualNetwork. The typical value for out_channels here is the number of classes.

Parameters:

• in_channels (int)

• out_channels (int)

• num_blocks (int)

• hidden_dim (int)

• data_dim (int)

• in_proj_cfg (LazyConfig)

• out_proj_cfg (LazyConfig)

• norm_cfg (LazyConfig)

• block_cfg (LazyConfig)

• dropout_in_cfg (LazyConfig)

• condition_in_proj_cfg (LazyConfig | None)

• target_size (int | Sequence[int] | None)

• gradient_checkpointing (bool)

forward(input_and_condition)

Run classification forward pass: backbone → GAP → norm → projection.

Parameters:

input_and_condition (dict[str, Tensor]) –

Dictionary with two keys:

• "input" — signal tensor of shape [B, *spatial, in_channels].

• "condition" — optional conditioning tensor of shape [B, *spatial_cond, hidden_dim], or None.

Returns:

Single-key dict:

• "logits" — shape [B, out_channels] (one logit vector per sample, all spatial information collapsed by global average pooling).

Return type:

dict[str, torch.Tensor]