drivingModel.py

import torch
import torch.nn as nn
import torch.nn.functional as F

# -----------------------------
# BasicBlock (ResNet18/34)
# -----------------------------
class BasicBlock(nn.Module):
    # BasicBlock는 채널 확장 없음
    expansion = 1

    def __init__(self, in_channels, out_channels, stride=1, downsample=None):
        super(BasicBlock, self).__init__()

        # 첫 번째 3x3 conv
        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channels)

        # 두 번째 3x3 conv
        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, stride=1, padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(out_channels)

        # identity(스킵) 변환기가 필요하면 전달
        self.downsample = downsample
        self.relu = nn.ReLU(inplace=True)
    
    def forward(self, x):
        # 스킵 경로용 원본 입략
        identity = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)

        # 필요하면 identity를 downsample로 조정(1x1 conv + BN)
        if self.downsample is not None:
            identity = self.downsample(x)

        # Element-wise Addition (요소별 합)
        # F(x) + x
        out += identity
        out = self.relu(out)
        return out

class DribingResNet(nn.Module):
    def __init__(self, num_classes=3):
        super(DribingResNet, self).__init__()
        block = BasicBlock

        # ResNet18
        layers = [2, 2, 2, 2]

        self.in_channels = 64
        # 초기 입력 처리 (Initial Conv Layer)
        # 보통 7x7 conv를 쓰지만 CIFAR-10 정도의 작은 이미지엔 3x3을 쓰기도 함
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)

        # 공간 해상도 감소, 모델의 연산량 감소
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)

        # Residual Layers (4개의 스테이지)
        # _make_layer 함수를 통해 블록을 반복해서 쌓음
        self.layer1 = self._make_layer(block, 64, layers[0], stride=1)
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
    
        # 최종 출력물 (Avrage Pooling + Fully Connected)
        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(512 * block.expansion, num_classes)

        # 파라미터들의 초기값을 설정
        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.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

    def _make_layer(self, block, out_channels, blocks, stride):
        downsample = None

        if stride != 1 or self.in_channels != out_channels * block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.in_channels, out_channels * block.expansion, kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(out_channels * block.expansion),
            )

        layers = []

        # 첫 블록: stride 가능 (downsample 포함)
        layers.append(block(self.in_channels, out_channels, stride, downsample))
        self.in_channels = out_channels * block.expansion

        # 남은 블록들: stride=1 (feature map size 유지)
        # 첫 블록은 상단에서 생성 했으니 반복문은 1부터 시작
        for _ in range(1, blocks):
            layers.append(block(self.in_channels, out_channels))

        return nn.Sequential(*layers)
    
    def forward(self, x):
        # 초기 변환
        out = F.relu(self.bn1(self.conv1(x)))
        out = self.maxpool(out)

        # Residual Blocks 통과
        out = self.layer1(out)
        out = self.layer2(out)
        out = self.layer3(out)
        out = self.layer4(out)

        # 분류기
        out = self.avgpool(out)
        out = torch.flatten(out, 1)
        out = self.fc(out)

        return out