RepLKNet¶

class mmpretrain.models.backbones.RepLKNet(arch, in_channels=3, ffn_ratio=4, out_indices=(3,), strides=(2, 2, 2, 2), dilations=(1, 1, 1, 1), frozen_stages=-1, conv_cfg=None, norm_cfg={'type': 'BN'}, act_cfg={'type': 'ReLU'}, with_cp=False, drop_path_rate=0.3, small_kernel_merged=False, norm_intermediate_features=False, norm_eval=False, init_cfg=[{'type': 'Kaiming', 'layer': ['Conv2d']}, {'type': 'Constant', 'val': 1, 'layer': ['_BatchNorm', 'GroupNorm']}])[source]¶

RepLKNet backbone.

A PyTorch impl of : Scaling Up Your Kernels to 31x31: Revisiting Large Kernel Design in CNNs

Parameters:

arch (str | dict) –
The parameter of RepLKNet. If it’s a dict, it should contain the following keys:
- large_kernel_sizes (Sequence[int]):
  Large kernel size in each stage.
- layers (Sequence[int]): Number of blocks in each stage.
- channels (Sequence[int]): Number of channels in each stage.
- small_kernel (int): size of the parallel small kernel.
- dw_ratio (float): The intermediate channels
  expansion ratio of the block.
in_channels (int) – Number of input image channels. Default to 3.
ffn_ratio (float) – Mlp expansion ratio. Defaults to 4.
out_indices (Sequence[int]) – Output from which stages. Default to (3, ).
strides (Sequence[int]) – Strides of the first block of each stage. Default to (2, 2, 2, 2).
dilations (Sequence[int]) – Dilation of each stage. Default to (1, 1, 1, 1).
frozen_stages (int) – Stages to be frozen (all param fixed). -1 means not freezing any parameters. Default to -1.
conv_cfg (dict | None) – The config dict for conv layers. Default to None.
norm_cfg (dict) – The config dict for norm layers. Default to dict(type='BN').
act_cfg (dict) – Config dict for activation layer. Default to dict(type='ReLU').
with_cp (bool) – Use checkpoint or not. Using checkpoint will save some memory while slowing down the training speed. Default to False.
deploy (bool) – Whether to switch the model structure to deployment mode. Default to False.
norm_intermediate_features (bool) – Construct and config norm layer or not. Using True will normalize the intermediate features for downstream dense prediction tasks.
norm_eval (bool) – Whether to set norm layers to eval mode, namely, freeze running stats (mean and var). Note: Effect on Batch Norm and its variants only. Default to False.
init_cfg (dict or list[dict], optional) – Initialization config dict.