"""
Ported to pytorch thanks to [tstandley](https://github.com/tstandley/Xception-PyTorch)

@author: tstandley
Adapted by cadene

Creates an Xception Model as defined in:

Francois Chollet
Xception: Deep Learning with Depthwise Separable Convolutions
https://arxiv.org/pdf/1610.02357.pdf

This weights ported from the Keras implementation. Achieves the following performance on the validation set:

Loss:0.9173 Prec@1:78.892 Prec@5:94.292

REMEMBER to set your image size to 3x299x299 for both test and validation

normalize = transforms.Normalize(mean=[0.5, 0.5, 0.5],
                                  std=[0.5, 0.5, 0.5])

The resize parameter of the validation transform should be 333, and make sure to center crop at 299x299
"""
import torch.nn as nn
import torch.nn.functional as F
from utils.Stomach_Config import Model_Config

from timm.layers import create_classifier
from .ViT_Model import ViTBranch  # 添加或替換這個導入


class SeparableConv2d(nn.Module):
    def __init__(
            self,
            in_channels: int,
            out_channels: int,
            kernel_size: int = 1,
            stride: int = 1,
            padding: int = 0,
            dilation: int = 1,
            device=None,
            dtype=None,
    ):
        dd = {'device': device, 'dtype': dtype}
        super().__init__()

        self.conv1 = nn.Conv2d(
            in_channels,
            in_channels,
            kernel_size,
            stride,
            padding,
            dilation,
            groups=in_channels,
            bias=False,
            **dd,
        )
        self.pointwise = nn.Conv2d(in_channels, out_channels, 1, 1, 0, 1, 1, bias=False, **dd)

    def forward(self, x):
        x = self.conv1(x)
        x = self.pointwise(x)
        return x


class Block(nn.Module):
    def __init__(
            self,
            in_channels: int,
            out_channels: int,
            reps: int,
            strides: int = 1,
            start_with_relu: bool = True,
            grow_first: bool = True,
            device=None,
            dtype=None,
    ):
        dd = {'device': device, 'dtype': dtype}
        super().__init__()

        if out_channels != in_channels or strides != 1:
            self.skip = nn.Conv2d(in_channels, out_channels, 1, stride=strides, bias=False, **dd)
            self.skipbn = nn.BatchNorm2d(out_channels, **dd)
        else:
            self.skip = None

        rep = []
        for i in range(reps):
            if grow_first:
                inc = in_channels if i == 0 else out_channels
                outc = out_channels
            else:
                inc = in_channels
                outc = in_channels if i < (reps - 1) else out_channels
            rep.append(nn.ReLU(inplace=True))
            rep.append(SeparableConv2d(inc, outc, 3, stride=1, padding=1, **dd))
            rep.append(nn.BatchNorm2d(outc, **dd))

        if not start_with_relu:
            rep = rep[1:]
        else:
            rep[0] = nn.ReLU(inplace=False)

        if strides != 1:
            rep.append(nn.MaxPool2d(3, strides, 1))
        self.rep = nn.Sequential(*rep)

    def forward(self, inp):
        x = self.rep(inp)

        if self.skip is not None:
            skip = self.skip(inp)
            skip = self.skipbn(skip)
        else:
            skip = inp

        x += skip
        return x


class Xception(nn.Module):
    """
    Xception optimized for the ImageNet dataset, as specified in
    https://arxiv.org/pdf/1610.02357.pdf
    """

    def __init__(
            self,
            num_classes: int = 1000,
            in_chans: int = 3,
            drop_rate: float = 0.,
            global_pool: str = 'avg',
            device=None,
            dtype=None,
    ):
        """ Constructor
        Args:
            num_classes: number of classes
        """
        super().__init__()
        dd = {'device': device, 'dtype': dtype}
        self.drop_rate = drop_rate
        self.global_pool = global_pool
        self.num_classes = num_classes
        self.num_features = self.head_hidden_size = 2048

        self.conv1 = nn.Conv2d(in_chans, 32, 3, 2, 0, bias=False, **dd)
        self.bn1 = nn.BatchNorm2d(32, **dd)
        self.act1 = nn.ReLU(inplace=True)

        self.conv2 = nn.Conv2d(32, 64, 3, bias=False, **dd)
        self.bn2 = nn.BatchNorm2d(64, **dd)
        self.act2 = nn.ReLU(inplace=True)

        self.block1 = Block(64, 128, 2, 2, start_with_relu=False, **dd)
        self.block2 = Block(128, 256, 2, 2, **dd)
        self.block3 = Block(256, 728, 2, 2, **dd)

        self.block4 = Block(728, 728, 3, 1, **dd)
        self.block5 = Block(728, 728, 3, 1, **dd)
        self.block6 = Block(728, 728, 3, 1, **dd)
        self.block7 = Block(728, 728, 3, 1, **dd)

        self.block8 = Block(728, 728, 3, 1, **dd)
        self.block9 = Block(728, 728, 3, 1, **dd)
        self.block10 = Block(728, 728, 3, 1, **dd)
        self.block11 = Block(728, 728, 3, 1, **dd)

        self.block12 = Block(728, 1024, 2, 2, grow_first=False, **dd)

        self.conv3 = SeparableConv2d(1024, 1536, 3, 1, 1, **dd)
        self.bn3 = nn.BatchNorm2d(1536, **dd)
        self.act3 = nn.ReLU(inplace=True)

        self.conv4 = SeparableConv2d(1536, self.num_features, 3, 1, 1, **dd)
        self.bn4 = nn.BatchNorm2d(self.num_features, **dd)
        self.act4 = nn.ReLU(inplace=True)
        self.feature_info = [
            dict(num_chs=64, reduction=2, module='act2'),
            dict(num_chs=128, reduction=4, module='block2.rep.0'),
            dict(num_chs=256, reduction=8, module='block3.rep.0'),
            dict(num_chs=728, reduction=16, module='block12.rep.0'),
            dict(num_chs=2048, reduction=32, module='act4'),
        ]

        self.global_pool, self.fc = create_classifier(self.num_features, self.num_classes, pool_type=global_pool, **dd)
        self.hidden_layer = nn.Linear(2048, Model_Config["Linear Hidden Nodes"])  # 隱藏層，輸入大小取決於 Xception 的輸出大小
        self.output_layer = nn.Linear(Model_Config["Linear Hidden Nodes"], Model_Config["Output Linear Nodes"])  # 輸出層，依據分類數目設定
        
        # 激活函數與 dropout
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(Model_Config["Dropout Rate"])

        # #------- init weights --------
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isinstance(m, nn.BatchNorm2d):
                m.weight.data.fill_(1)
                m.bias.data.zero_()

    def reset_classifier(self, num_classes: int, global_pool: str = 'avg'):
        self.num_classes = num_classes
        self.global_pool, self.fc = create_classifier(self.num_features, self.num_classes, pool_type=global_pool)

    def forward_features(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.act1(x)

        x = self.conv2(x)
        x = self.bn2(x)
        x = self.act2(x)

        x = self.block1(x)
        x = self.block2(x)
        x = self.block3(x)
        x = self.block4(x)
        x = self.block5(x)
        x = self.block6(x)
        x = self.block7(x)
        x = self.block8(x)
        x = self.block9(x)
        x = self.block10(x)
        x = self.block11(x)
        x = self.block12(x)

        x = self.conv3(x)
        x = self.bn3(x)
        x = self.act3(x)

        x = self.conv4(x)
        x = self.bn4(x)
        x = self.act4(x)
        return x

    def forward_head(self, x):
        x = self.global_pool(x)
        return x

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

        x = self.dropout(x)                   # Dropout
        x = self.hidden_layer(x)
        x = self.relu(x)   # 隱藏層 + ReLU
        x = self.output_layer(x)              # 輸出層
        return x


# xception_vit.py
class XceptionWithViT(nn.Module):
    """
    Xception + ViT Branch (Middle Flow 後融合)
    """
    def __init__(
        self,
        num_classes: int = 1000,
        in_chans: int = 3,
        drop_rate: float = 0.0,
        global_pool: str = 'avg',
        vit_patch_size: int = 4,
        vit_depth: int = 3,
        vit_heads: int = 8,
        device=None,
        dtype=None,
    ):
        super().__init__()
        dd = {'device': device, 'dtype': dtype}
        self.drop_rate = drop_rate
        self.global_pool = global_pool
        self.num_features = 2048

        # === Entry Flow ===
        self.conv1 = nn.Conv2d(in_chans, 32, 3, 2, 0, bias=False, **dd)
        self.bn1 = nn.BatchNorm2d(32, **dd)
        self.act1 = nn.ReLU(inplace=True)

        self.conv2 = nn.Conv2d(32, 64, 3, bias=False, **dd)
        self.bn2 = nn.BatchNorm2d(64, **dd)
        self.act2 = nn.ReLU(inplace=True)

        self.block1 = Block(64, 128, 2, 2, start_with_relu=False, **dd)
        self.block2 = Block(128, 256, 2, 2, **dd)
        self.block3 = Block(256, 728, 2, 2, **dd)

        # === Middle Flow ===
        for i in range(4, 12):
            setattr(self, f'block{i}', Block(728, 728, 3, 1, **dd))

        # === ViT Branch ===
        self.vit_branch = ViTBranch(
            in_chs=728,
            embed_dim=728,
            patch_size=4,
            depth=3,
            num_heads=8,
            mlp_ratio=4.0,
            drop_rate=0.0,
            attn_drop_rate=0.0,
            device=device,
            dtype=dtype
        )

        # === Exit Flow ===
        self.block12 = Block(728, 1024, 2, 2, grow_first=False, **dd)
        self.conv3 = SeparableConv2d(1024, 1536, 3, 1, 1, **dd)
        self.bn3 = nn.BatchNorm2d(1536, **dd)
        self.act3 = nn.ReLU(inplace=True)
        self.conv4 = SeparableConv2d(1536, self.num_features, 3, 1, 1, **dd)
        self.bn4 = nn.BatchNorm2d(self.num_features, **dd)
        self.act4 = nn.ReLU(inplace=True)

        # === Classifier ===
        self.global_pool, _ = create_classifier(self.num_features, num_classes, pool_type=global_pool, **dd)
        self.hidden_layer = nn.Linear(2048, Model_Config["Linear Hidden Nodes"])
        self.output_layer = nn.Linear(Model_Config["Linear Hidden Nodes"], Model_Config["Output Linear Nodes"])
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(Model_Config["Dropout Rate"])

        # === Weight Init ===
        self._init_weights()

    def _init_weights(self):
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isinstance(m, (nn.BatchNorm2d, nn.LayerNorm)):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)
            elif isinstance(m, nn.Linear):
                nn.init.normal_(m.weight, 0, 0.01)
                if m.bias is not None:
                    nn.init.constant_(m.bias, 0)

    def forward_features(self, x):
        # Entry
        x = self.act1(self.bn1(self.conv1(x)))
        x = self.act2(self.bn2(self.conv2(x)))
        x = self.block1(x)
        x = self.block2(x)
        x = self.block3(x)

        # Middle Flow
        for i in range(4, 12):
            x = getattr(self, f'block{i}')(x)

        # === ViT Fusion ===
        vit_out = self.vit_branch(x)
        x = x + vit_out  # element-wise add

        # Exit Flow
        x = self.block12(x)
        x = self.act3(self.bn3(self.conv3(x)))
        x = self.act4(self.bn4(self.conv4(x)))
        return x

    def forward(self, x):
        x = self.forward_features(x)
        x = self.global_pool(x)
        x = self.dropout(x)
        x = self.relu(self.hidden_layer(x))
        x = self.output_layer(x)
        return x