spandrel.architectures.HAT.HAT

class documentation

class HAT(nn.Module): (source)

Constructor: HAT(img_size, patch_size, in_chans, embed_dim, ...)

View In Hierarchy

Hybrid Attention Transformer
A PyTorch implementation of : Activating More Pixels in Image Super-Resolution Transformer. Some codes are based on SwinIR.

Parameters
img_sizeInput image size. Default 64
patch_sizePatch size. Default: 1
in_chansNumber of input image channels. Default: 3
embed_dimPatch embedding dimension. Default: 96
depthsDepth of each Swin Transformer layer.
num_headsNumber of attention heads in different layers.
window_sizeWindow size. Default: 7
mlp_ratioRatio of mlp hidden dim to embedding dim. Default: 4
qkv_biasIf True, add a learnable bias to query, key, value. Default: True
qk_scaleOverride default qk scale of head_dim ** -0.5 if set. Default: None
drop_rateDropout rate. Default: 0
attn_drop_rateAttention dropout rate. Default: 0
drop_path_rateStochastic depth rate. Default: 0.1
norm_layerNormalization layer. Default: nn.LayerNorm.
apeIf True, add absolute position embedding to the patch embedding. Default: False
patch_normIf True, add normalization after patch embedding. Default: True
use_checkpointWhether to use checkpointing to save memory. Default: False
upscaleUpscale factor. 2/3/4/8 for image SR, 1 for denoising and compress artifact reduction
img_rangeImage range. 1. or 255.
upsamplerThe reconstruction reconstruction module. 'pixelshuffle'/'pixelshuffledirect'/'nearest+conv'/None
resi_connectionThe convolutional block before residual connection. '1conv'/'3conv'
Method __init__ Undocumented
Method calculate_mask Undocumented
Method calculate_rpi_oca Undocumented
Method calculate_rpi_sa Undocumented
Method check_image_size Undocumented
Method forward Undocumented
Method forward_features Undocumented
Class Variable hyperparameters Undocumented
Instance Variable absolute_pos_embed Undocumented
Instance Variable ape Undocumented
Instance Variable conv_after_body Undocumented
Instance Variable conv_before_upsample Undocumented
Instance Variable conv_first Undocumented
Instance Variable conv_last Undocumented
Instance Variable embed_dim Undocumented
Instance Variable img_range Undocumented
Instance Variable layers Undocumented
Instance Variable mean Undocumented
Instance Variable mlp_ratio Undocumented
Instance Variable norm Undocumented
Instance Variable num_features Undocumented
Instance Variable num_layers Undocumented
Instance Variable overlap_ratio Undocumented
Instance Variable patch_embed Undocumented
Instance Variable patch_norm Undocumented
Instance Variable patch_unembed Undocumented
Instance Variable patches_resolution Undocumented
Instance Variable pos_drop Undocumented
Instance Variable shift_size Undocumented
Instance Variable upsample Undocumented
Instance Variable upsampler Undocumented
Instance Variable upscale Undocumented
Instance Variable window_size Undocumented
Method _init_weights Undocumented
def __init__(self, *, img_size=64, patch_size=1, in_chans=3, embed_dim=96, depths=(6, 6, 6, 6), num_heads=(6, 6, 6, 6), window_size=7, compress_ratio: int|float = 3, squeeze_factor: int|float = 30, conv_scale=0.01, overlap_ratio=0.5, mlp_ratio=4.0, qkv_bias=True, qk_scale=None, drop_rate=0.0, attn_drop_rate=0.0, drop_path_rate=0.1, norm_layer=nn.LayerNorm, ape=False, patch_norm=True, use_checkpoint=False, upscale=1, img_range=1.0, upsampler: Literal['pixelshuffle'] = 'pixelshuffle', resi_connection='1conv', num_feat=64): (source)

Undocumented

def calculate_mask(self, x_size): (source)

Undocumented

def calculate_rpi_oca(self): (source)

Undocumented

def calculate_rpi_sa(self): (source)

Undocumented

def check_image_size(self, x): (source)

Undocumented

def forward(self, x): (source)

Undocumented

def forward_features(self, x): (source)

Undocumented

hyperparameters: dict = (source)

Undocumented

absolute_pos_embed = (source)

Undocumented

ape: bool = (source)

Undocumented

conv_after_body = (source)

Undocumented

conv_before_upsample = (source)

Undocumented

conv_first = (source)

Undocumented

conv_last = (source)

Undocumented

embed_dim: int = (source)

Undocumented

img_range = (source)

Undocumented

layers = (source)

Undocumented

mean = (source)

Undocumented

mlp_ratio: float = (source)

Undocumented

norm = (source)

Undocumented

num_features = (source)

Undocumented

num_layers = (source)

Undocumented

overlap_ratio = (source)

Undocumented

patch_embed = (source)

Undocumented

patch_norm: bool = (source)

Undocumented

patch_unembed = (source)

Undocumented

patches_resolution = (source)

Undocumented

pos_drop = (source)

Undocumented

shift_size = (source)

Undocumented

upsample = (source)

Undocumented

upsampler = (source)

Undocumented

upscale = (source)

Undocumented

window_size: int = (source)

Undocumented

def _init_weights(self, m): (source)

Undocumented