WiT_model.py

import torch
import torch.nn as nn
import torch.nn.functional
import torch.nn.functional as F
from functools import partial
from base_net import *

import timm
from timm.models.layers import DropPath, to_2tuple, trunc_normal_
import types
import math
from abc import ABCMeta, abstractmethod
from mmcv.cnn import ConvModule
import pdb

import pywt
import argparse
from torch.autograd import Variable
from torch_wavelet import DWT_2D, IDWT_2D


class EncoderTransformer(nn.Module):
    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dims=[64, 128, 256, 512],
                 num_heads=[1, 2, 4, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=False, qk_scale=None, drop_rate=0.,
                 attn_drop_rate=0., drop_path_rate=0., norm_layer=nn.LayerNorm,
                 depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1]):
        super().__init__()
        self.num_classes = num_classes
        self.depths = depths

        # patch embedding
        self.patch_embed1 = OverlapPatchEmbed(img_size=img_size, patch_size=7, stride=4, in_chans=in_chans,
                                              embed_dim=embed_dims[0])
        self.patch_embed2 = OverlapPatchEmbed(img_size=img_size // 4, patch_size=3, stride=2, in_chans=embed_dims[0],
                                              embed_dim=embed_dims[1])
        self.patch_embed3 = OverlapPatchEmbed(img_size=img_size // 8, patch_size=3, stride=2, in_chans=embed_dims[1],
                                              embed_dim=embed_dims[2])
        self.patch_embed4 = OverlapPatchEmbed(img_size=img_size // 16, patch_size=3, stride=2, in_chans=embed_dims[2],
                                              embed_dim=embed_dims[3])
        # for Intra-patch transformer blocks

        self.mini_patch_embed1 = OverlapPatchEmbed(img_size=img_size // 4, patch_size=3, stride=2, in_chans=embed_dims[0],
                          embed_dim=embed_dims[1])
        self.mini_patch_embed2 = OverlapPatchEmbed(img_size=img_size // 8, patch_size=3, stride=2, in_chans=embed_dims[1],
                                              embed_dim=embed_dims[2])
        self.mini_patch_embed3 = OverlapPatchEmbed(img_size=img_size // 16, patch_size=3, stride=2, in_chans=embed_dims[2],
                                              embed_dim=embed_dims[3])
        self.mini_patch_embed4 = OverlapPatchEmbed(img_size=img_size // 32, patch_size=3, stride=2, in_chans=embed_dims[0],
                                              embed_dim=embed_dims[3])

        ### Main encoder (M) ###
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule
        cur = 0
        self.block1 = nn.ModuleList([Block(
            dim=embed_dims[0], num_heads=num_heads[0], mlp_ratio=mlp_ratios[0], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[0])
            for i in range(depths[0])])
        self.norm1 = norm_layer(embed_dims[0])

        # mini-patch encoder (m) #
        self.patch_block1 = nn.ModuleList([Block(
            dim=embed_dims[1], num_heads=num_heads[0], mlp_ratio=mlp_ratios[0], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[0])
            for i in range(1)])
        self.pnorm1 = norm_layer(embed_dims[1])

        ### Main encoder (M) ###
        cur += depths[0]
        self.block2 = nn.ModuleList([Block(
            dim=embed_dims[1], num_heads=num_heads[1], mlp_ratio=mlp_ratios[1], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[1])
            for i in range(depths[1])])
        self.norm2 = norm_layer(embed_dims[1])

        # mini-patch encoder (m) #
        self.patch_block2 = nn.ModuleList([Block(
            dim=embed_dims[2], num_heads=num_heads[1], mlp_ratio=mlp_ratios[1], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[1])
            for i in range(1)])
        self.pnorm2 = norm_layer(embed_dims[2])

        ### Main encoder (M) ###
        cur += depths[1]
        self.block3 = nn.ModuleList([Block(
            dim=embed_dims[2], num_heads=num_heads[2], mlp_ratio=mlp_ratios[2], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[2])
            for i in range(depths[2])])
        self.norm3 = norm_layer(embed_dims[2])

        # mini-patch encoder (m) #
        self.patch_block3 = nn.ModuleList([Block(
            dim=embed_dims[3], num_heads=num_heads[1], mlp_ratio=mlp_ratios[2], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[2])
            for i in range(1)])
        self.pnorm3 = norm_layer(embed_dims[3])

        ### Main encoder (M) ###
        cur += depths[2]
        self.block4 = nn.ModuleList([Block(
            dim=embed_dims[3], num_heads=num_heads[3], mlp_ratio=mlp_ratios[3], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[3])
            for i in range(depths[3])])
        self.norm4 = norm_layer(embed_dims[3])

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def init_weights(self, pretrained=None):
        if isinstance(pretrained, str):
            logger = get_root_logger()
            load_checkpoint(self, pretrained, map_location='cpu', strict=False, logger=logger)

    def reset_drop_path(self, drop_path_rate):
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(self.depths))]
        cur = 0
        for i in range(self.depths[0]):
            self.block1[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[0]
        for i in range(self.depths[1]):
            self.block2[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[1]
        for i in range(self.depths[2]):
            self.block3[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[2]
        for i in range(self.depths[3]):
            self.block4[i].drop_path.drop_prob = dpr[cur + i]

    def forward_features(self, x):
        B = x.shape[0]
        outs = []
        embed_dims=[64, 128, 320, 512]

        # stage 1
        x1, H1, W1 = self.patch_embed1(x)
        x2, H2, W2 = self.mini_patch_embed1(x1.permute(0,2,1).reshape(B,embed_dims[0],H1,W1))

        for i, blk in enumerate(self.block1):
            x1 = blk(x1, H1, W1)
        x1 = self.norm1(x1)
        x1 = x1.reshape(B, H1, W1, -1).permute(0, 3, 1, 2).contiguous()

        for i, blk in enumerate(self.patch_block1):
            x2 = blk(x2, H2, W2)
        x2 = self.pnorm1(x2)
        x2 = x2.reshape(B, H2, W2, -1).permute(0, 3, 1, 2).contiguous()

        outs.append(x1)

        # stage 2
        x1, H1, W1 = self.patch_embed2(x1)
        x1 = x1.permute(0,2,1).reshape(B,embed_dims[1],H1,W1)+x2
        x2, H2, W2 = self.mini_patch_embed2(x1)

        x1 = x1.view(x1.shape[0],x1.shape[1],-1).permute(0,2,1)

        for i, blk in enumerate(self.block2):
            x1 = blk(x1, H1, W1)
        x1 = self.norm2(x1)
        x1 = x1.reshape(B, H1, W1, -1).permute(0, 3, 1, 2).contiguous()
        outs.append(x1)

        for i, blk in enumerate(self.patch_block2):
            x2 = blk(x2, H2, W2)
        x2 = self.pnorm2(x2)
        x2 = x2.reshape(B, H2, W2, -1).permute(0, 3, 1, 2).contiguous()


        # stage 3
        x1, H1, W1 = self.patch_embed3(x1)
        x1 = x1.permute(0,2,1).reshape(B,embed_dims[2],H1,W1)+x2
        x2, H2, W2 = self.mini_patch_embed3(x1)

        x1 = x1.view(x1.shape[0],x1.shape[1],-1).permute(0,2,1)

        for i, blk in enumerate(self.block3):
            x1 = blk(x1, H1, W1)
        x1 = self.norm3(x1)
        x1 = x1.reshape(B, H1, W1, -1).permute(0, 3, 1, 2).contiguous()
        outs.append(x1)

        for i, blk in enumerate(self.patch_block3):
            x2 = blk(x2, H2, W2)
        x2 = self.pnorm3(x2)
        x2 = x2.reshape(B, H2, W2, -1).permute(0, 3, 1, 2).contiguous()

        # stage 4
        x1, H1, W1 = self.patch_embed4(x1)
        x1 = x1.permute(0,2,1).reshape(B,embed_dims[3],H1,W1)+x2

        x1 = x1.view(x1.shape[0],x1.shape[1],-1).permute(0,2,1)

        for i, blk in enumerate(self.block4):
            x1 = blk(x1, H1, W1)
        x1 = self.norm4(x1)
        x1 = x1.reshape(B, H1, W1, -1).permute(0, 3, 1, 2).contiguous()
        outs.append(x1)

        return outs

    def forward(self, x):
        x = self.forward_features(x)

        return x

    """ 
    Image Patch Embedding
    """
class OverlapPatchEmbed(nn.Module):


    def __init__(self, img_size=224, patch_size=7, stride=4, in_chans=3, embed_dim=768):
        super().__init__()
        img_size = to_2tuple(img_size)
        patch_size = to_2tuple(patch_size)

        self.img_size = img_size
        self.patch_size = patch_size
        self.H, self.W = img_size[0] // patch_size[0], img_size[1] // patch_size[1]
        self.num_patches = self.H * self.W
        self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=stride,
                              padding=(patch_size[0] // 2, patch_size[1] // 2))
        self.norm = nn.LayerNorm(embed_dim)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x):
        # pdb.set_trace()
        x = self.proj(x)
        _, _, H, W = x.shape
        x = x.flatten(2).transpose(1, 2)
        x = self.norm(x)

        return x, H, W

def resize(input,
           size=None,
           scale_factor=None,
           mode='nearest',
           align_corners=None,
           warning=True):
    if warning:
        if size is not None and align_corners:
            input_h, input_w = tuple(int(x) for x in input.shape[2:])
            output_h, output_w = tuple(int(x) for x in size)
            if output_h > input_h or output_w > output_h:
                if ((output_h > 1 and output_w > 1 and input_h > 1
                     and input_w > 1) and (output_h - 1) % (input_h - 1)
                        and (output_w - 1) % (input_w - 1)):
                    warnings.warn(
                        f'When align_corners={align_corners}, '
                        'the output would more aligned if '
                        f'input size {(input_h, input_w)} is `x+1` and '
                        f'out size {(output_h, output_w)} is `nx+1`')
    return F.interpolate(input, size, scale_factor, mode, align_corners)

############################################################

class Mlp(nn.Module):
    def __init__(self, in_features, hidden_features=None, out_features=None, act_layer=nn.GELU, drop=0.):
        super().__init__()
        out_features = out_features or in_features
        hidden_features = hidden_features or in_features
        self.fc1 = nn.Linear(in_features, hidden_features)
        self.dwconv = DWConv(hidden_features)
        self.act = act_layer()
        self.fc2 = nn.Linear(hidden_features, out_features)
        self.drop = nn.Dropout(drop)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x, H, W):
        x = self.fc1(x)
        x = self.dwconv(x, H, W)
        x = self.act(x)
        x = self.drop(x)
        x = self.fc2(x)
        x = self.drop(x)
        return x


###### WiT-Wavelet Attention ######

class Attention(nn.Module):
    def __init__(self, dim, num_heads, sr_ratio, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0.):
        super().__init__()
        assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}."
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = head_dim**-0.5
        self.sr_ratio = sr_ratio

        self.dwt = DWT_2D(wave='haar')
        self.idwt = IDWT_2D(wave='haar')
        self.reduce = nn.Sequential(
            nn.Conv2d(dim, dim//4, kernel_size=1, padding=0, stride=1),
            nn.BatchNorm2d(dim//4),
            nn.ReLU(inplace=True),
        )
        self.filter = nn.Sequential(
            nn.Conv2d(dim, dim, kernel_size=3, padding=1, stride=1, groups=1),
            nn.BatchNorm2d(dim),
            nn.ReLU(inplace=True),
        )
        self.kv_embed = nn.Conv2d(dim, dim, kernel_size=sr_ratio, stride=sr_ratio) if sr_ratio > 1 else nn.Identity()
        self.q = nn.Linear(dim, dim, bias=qkv_bias)
        self.kv = nn.Sequential(
            nn.LayerNorm(dim),
            nn.Linear(dim, dim * 2, bias=qkv_bias)
        )
        self.proj = nn.Linear(dim+dim//4, dim)
        self.proj_drop = nn.Dropout(proj_drop)
        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()
                
###### DWT call ##############
    
    def forward(self, x, H, W):
        B, N, C = x.shape
        q = self.q(x).reshape(B, N, self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)

        x = x.view(B, H, W, C).permute(0, 3, 1, 2) #(BHWC)
        x_dwt = self.dwt(self.reduce(x))
        x_dwt = self.filter(x_dwt)
        
###### iDWT call ##############
        
        x_idwt = self.idwt(x_dwt)
        x_idwt = x_idwt.view(B, -1, x_idwt.size(-2)*x_idwt.size(-1)).transpose(1, 2)

        kv = self.kv_embed(x_dwt).reshape(B, C, -1).permute(0, 2, 1)
        kv = self.kv(kv).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        k, v = kv[0], kv[1]

        attn = (q @ k.transpose(-2, -1)) * self.scale
        attn = attn.softmax(dim=-1)
        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        
###### Conditional statement if tensor mismatch happens ########
        
        if x.shape != x_idwt.shape:
            B,N,C = x_idwt.shape
            diff = abs(x.shape[1]-x_idwt.shape[1])
            add_tensor = torch.rand((B, diff, C)).cuda()
            x_idwt = torch.cat((x_idwt, add_tensor), 1)

        x = self.proj(torch.cat([x, x_idwt], dim=2))
        #x = self.proj(x)

        return x
        
###### DWT-MHSA ENDS-HERE @ WiT-Attention ######

class Attention_dec(nn.Module):
    def __init__(self, dim, num_heads=8, qkv_bias=False, qk_scale=None, attn_drop=0., proj_drop=0., sr_ratio=1):
        super().__init__()
        assert dim % num_heads == 0, f"dim {dim} should be divided by num_heads {num_heads}."

        self.dim = dim
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5

        self.q = nn.Linear(dim, dim, bias=qkv_bias)
        self.kv = nn.Linear(dim, dim * 2, bias=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)

        self.task_query = nn.Parameter(torch.randn(1,48,dim))
        self.sr_ratio = sr_ratio
        if sr_ratio > 1:
            self.sr = nn.Conv2d(dim, dim, kernel_size=sr_ratio, stride=sr_ratio)
            self.norm = nn.LayerNorm(dim)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x, H, W):

        B, N, C = x.shape
        task_q = self.task_query

        ### fix the task parameters dimension
        if B>1:

            task_q = task_q.unsqueeze(0).repeat(B,1,1,1)
            task_q = task_q.squeeze(1)

        q = self.q(task_q).reshape(B, task_q.shape[1], self.num_heads, C // self.num_heads).permute(0, 2, 1, 3)

        if self.sr_ratio > 1:
            x_ = x.permute(0, 2, 1).reshape(B, C, H, W)
            x_ = self.sr(x_).reshape(B, C, -1).permute(0, 2, 1)
            x_ = self.norm(x_)
            kv = self.kv(x_).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        else:
            kv = self.kv(x).reshape(B, -1, 2, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4)
        k, v = kv[0], kv[1]
        q = torch.nn.functional.interpolate(q,size= (v.shape[2],v.shape[3]))
        attn = (q @ k.transpose(-2, -1)) * self.scale
        attn = attn.softmax(dim=-1)
        attn = self.attn_drop(attn)
        x = (attn @ v).transpose(1, 2).reshape(B, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)

        return x


class Block_dec(nn.Module):

    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, sr_ratio=1):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention_dec(
            dim,
            num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,
            attn_drop=attn_drop, proj_drop=drop, sr_ratio=sr_ratio)

        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x, H, W):

        x = x + self.drop_path(self.attn(self.norm1(x), H, W))
        x = x + self.drop_path(self.mlp(self.norm2(x), H, W))

        return x


class Block(nn.Module):

    def __init__(self, dim, num_heads, mlp_ratio=4., qkv_bias=False, qk_scale=None, drop=0., attn_drop=0.,
                 drop_path=0., act_layer=nn.GELU, norm_layer=nn.LayerNorm, sr_ratio=1):
        super().__init__()
        self.norm1 = norm_layer(dim)
        self.attn = Attention(
            dim,
            num_heads=num_heads, qkv_bias=qkv_bias, qk_scale=qk_scale,
            attn_drop=attn_drop, proj_drop=drop, sr_ratio=sr_ratio)

        self.drop_path = DropPath(drop_path) if drop_path > 0. else nn.Identity()
        self.norm2 = norm_layer(dim)
        mlp_hidden_dim = int(dim * mlp_ratio)
        self.mlp = Mlp(in_features=dim, hidden_features=mlp_hidden_dim, act_layer=act_layer, drop=drop)

        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def forward(self, x, H, W):

        x = x + self.drop_path(self.attn(self.norm1(x), H, W))
        x = x + self.drop_path(self.mlp(self.norm2(x), H, W))
        return x


class DWConv(nn.Module):
    def __init__(self, dim=768):
        super(DWConv, self).__init__()
        self.dwconv = nn.Conv2d(dim, dim, 3, 1, 1, bias=True, groups=dim)

    def forward(self, x, H, W):
        B, N, C = x.shape
        x = x.transpose(1, 2).view(B, C, H, W)
        x = self.dwconv(x)
        x = x.flatten(2).transpose(1, 2)

        return x

class DecoderTransformer(nn.Module):
    def __init__(self, img_size=224, patch_size=16, in_chans=3, num_classes=1000, embed_dims=[64, 128, 256, 512],
                 num_heads=[1, 2, 4, 8], mlp_ratios=[4, 4, 4, 4], qkv_bias=False, qk_scale=None, drop_rate=0.,
                 attn_drop_rate=0., drop_path_rate=0., norm_layer=nn.LayerNorm,
                 depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1]):
        super().__init__()
        self.num_classes = num_classes
        self.depths = depths

        # patch_embed
        self.patch_embed1 = OverlapPatchEmbed(img_size=img_size//16, patch_size=3, stride=2, in_chans=embed_dims[3],
                                              embed_dim=embed_dims[3])

        # transformer decoder
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(depths))]  # stochastic depth decay rule
        cur = 0
        self.block1 = nn.ModuleList([Block_dec(
            dim=embed_dims[3], num_heads=num_heads[3], mlp_ratio=mlp_ratios[3], qkv_bias=qkv_bias, qk_scale=qk_scale,
            drop=drop_rate, attn_drop=attn_drop_rate, drop_path=dpr[cur + i], norm_layer=norm_layer,
            sr_ratio=sr_ratios[3])
            for i in range(depths[0])])
        self.norm1 = norm_layer(embed_dims[3])

        cur += depths[0]


        self.apply(self._init_weights)

    def _init_weights(self, m):
        if isinstance(m, nn.Linear):
            trunc_normal_(m.weight, std=.02)
            if isinstance(m, nn.Linear) and m.bias is not None:
                nn.init.constant_(m.bias, 0)
        elif isinstance(m, nn.LayerNorm):
            nn.init.constant_(m.bias, 0)
            nn.init.constant_(m.weight, 1.0)
        elif isinstance(m, nn.Conv2d):
            fan_out = m.kernel_size[0] * m.kernel_size[1] * m.out_channels
            fan_out //= m.groups
            m.weight.data.normal_(0, math.sqrt(2.0 / fan_out))
            if m.bias is not None:
                m.bias.data.zero_()

    def init_weights(self, pretrained=None):
        if isinstance(pretrained, str):
            logger = get_root_logger()
            load_checkpoint(self, pretrained, map_location='cpu', strict=False, logger=logger)

    def reset_drop_path(self, drop_path_rate):
        dpr = [x.item() for x in torch.linspace(0, drop_path_rate, sum(self.depths))]
        cur = 0
        for i in range(self.depths[0]):
            self.block1[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[0]
        for i in range(self.depths[1]):
            self.block2[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[1]
        for i in range(self.depths[2]):
            self.block3[i].drop_path.drop_prob = dpr[cur + i]

        cur += self.depths[2]
        for i in range(self.depths[3]):
            self.block4[i].drop_path.drop_prob = dpr[cur + i]


    def forward_features(self, x):
        x=x[3]
        B = x.shape[0]
        outs = []

        # stage 1
        x, H, W = self.patch_embed1(x)
        for i, blk in enumerate(self.block1):
            x = blk(x, H, W)
        x = self.norm1(x)
        x = x.reshape(B, H, W, -1).permute(0, 3, 1, 2).contiguous()
        outs.append(x)

        return outs

    def forward(self, x):
        x = self.forward_features(x)
        # x = self.head(x)

        return x

class Tenc(EncoderTransformer):
    def __init__(self, **kwargs):
        super(Tenc, self).__init__(
            patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 4, 4], mlp_ratios=[2, 2, 2, 2],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[2, 2, 2, 2], sr_ratios=[4, 2, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1)

class Tdec(DecoderTransformer):
    def __init__(self, **kwargs):
        super(Tdec, self).__init__(
            patch_size=4, embed_dims=[64, 128, 320, 512], num_heads=[1, 2, 5, 8], mlp_ratios=[4, 4, 4, 4],
            qkv_bias=True, norm_layer=partial(nn.LayerNorm, eps=1e-6), depths=[3, 4, 6, 3], sr_ratios=[8, 4, 2, 1],
            drop_rate=0.0, drop_path_rate=0.1)


class convprojection(nn.Module):
    def __init__(self, path=None, **kwargs):
        super(convprojection,self).__init__()

        self.convd32x = UpsampleConvLayer(512, 512, kernel_size=4, stride=2)
        self.convd16x = UpsampleConvLayer(512, 320, kernel_size=4, stride=2)
        self.dense_4 = nn.Sequential(ResidualBlock(320))
        self.convd8x = UpsampleConvLayer(320, 128, kernel_size=4, stride=2)
        self.dense_3 = nn.Sequential(ResidualBlock(128))
        self.convd4x = UpsampleConvLayer(128, 64, kernel_size=4, stride=2)
        self.dense_2 = nn.Sequential(ResidualBlock(64))
        self.convd2x = UpsampleConvLayer(64, 16, kernel_size=4, stride=2)
        self.dense_1 = nn.Sequential( ResidualBlock(16))
        self.convd1x = UpsampleConvLayer(16, 8, kernel_size=4, stride=2)
        self.conv_output = ConvLayer(8, 3, kernel_size=3, stride=1, padding=1)

        self.active = nn.Tanh()

    def forward(self,x1,x2):

        res32x = self.convd32x(x2[0])

        if x1[3].shape[3] != res32x.shape[3] and x1[3].shape[2] != res32x.shape[2]:
            p2d = (0,-1,0,-1)
            res32x = F.pad(res32x,p2d,"constant",0)

        elif x1[3].shape[3] != res32x.shape[3] and x1[3].shape[2] == res32x.shape[2]:
            p2d = (0,-1,0,0)
            res32x = F.pad(res32x,p2d,"constant",0)
        elif x1[3].shape[3] == res32x.shape[3] and x1[3].shape[2] != res32x.shape[2]:
            p2d = (0,0,0,-1)
            res32x = F.pad(res32x,p2d,"constant",0)

        res16x = res32x + x1[3]
        res16x = self.convd16x(res16x)

        if x1[2].shape[3] != res16x.shape[3] and x1[2].shape[2] != res16x.shape[2]:
            p2d = (0,-1,0,-1)
            res16x = F.pad(res16x,p2d,"constant",0)
        elif x1[2].shape[3] != res16x.shape[3] and x1[2].shape[2] == res16x.shape[2]:
            p2d = (0,-1,0,0)
            res16x = F.pad(res16x,p2d,"constant",0)
        elif x1[2].shape[3] == res16x.shape[3] and x1[2].shape[2] != res16x.shape[2]:
            p2d = (0,0,0,-1)
            res16x = F.pad(res16x,p2d,"constant",0)

        res8x = self.dense_4(res16x) + x1[2]
        res8x = self.convd8x(res8x)
        res4x = self.dense_3(res8x) + x1[1]
        res4x = self.convd4x(res4x)
        res2x = self.dense_2(res4x) + x1[0]
        res2x = self.convd2x(res2x)
        x = res2x
        x = self.dense_1(x)
        x = self.convd1x(x)

        return x




### The network fine-tuning for snow datasets ###

class WiT(nn.Module):

    def __init__(self, path=None, **kwargs):
        super(WiT, self).__init__()

        self.Tenc = Tenc()

        self.Tdec = Tdec()

        self.convtail = convprojection()

        self.clean = ConvLayer(8, 3, kernel_size=3, stride=1, padding=1)

        self.active = nn.Tanh()

        if path is not None:
            self.load(path)

    def forward(self, x):

        x1 = self.Tenc(x)

        x2 = self.Tdec(x1)

        x = self.convtail(x1,x2)

        clean = self.active(self.clean(x))

        return clean


    def load(self, path):
        """
        Load checkpoint.
        """
        checkpoint = torch.load(path, map_location=lambda storage, loc: storage)
        model_state_dict_keys = self.state_dict().keys()
        checkpoint_state_dict_noprefix = strip_prefix_if_present(checkpoint['state_dict'], "module.")
        self.load_state_dict(checkpoint_state_dict_noprefix, strict=False)
        del checkpoint
        torch.cuda.empty_cache()