PyTorch U-Net模型转ONNX时固定宽度、可变高度输入的适配问题

阿华AIGC实验室

2026-4-14

我用PyTorch实现了一个用于PNG图像分割标注的U-Net模型，输入图像的宽度固定为512像素或其倍数，但高度范围在400到900像素之间。PyTorch模型（*.pth文件）运行完全正常，但在将其转换为ONNX格式时遇到了适配问题。

以下是我的模型转换代码：

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

# pip install torch onnx


class UNet(nn.Module):
    def __init__(self, n_channels, n_classes, bilinear=True):
        super().__init__()
        self.n_channels = n_channels
        self.n_classes = n_classes
        self.bilinear = bilinear

        self.inc = DoubleConv(n_channels, 64)
        self.down1 = Down(64, 128)
        self.down2 = Down(128, 256)
        self.down3 = Down(256, 512)
        factor = 2 if bilinear else 1
        self.down4 = Down(512, 1024 // factor)
        self.up1 = Up(1024, 512, bilinear)
        self.up2 = Up(512, 256, bilinear)
        self.up3 = Up(256, 128, bilinear)
        self.up4 = Up(128, 64 * factor, bilinear)
        self.outc = OutConv(64, n_classes)

    def forward(self, x):
        x1 = self.inc(x)
        x2 = self.down1(x1)
        x3 = self.down2(x2)
        x4 = self.down3(x3)
        x5 = self.down4(x4)
        x = self.up1(x5, x4)
        x = self.up2(x, x3)
        x = self.up3(x, x2)
        x = self.up4(x, x1)
        logits = self.outc(x)
        return logits


class DoubleConv(nn.Module):
    """(convolution => [BN] => ReLU) * 2"""

    def __init__(self, in_channels, out_channels, mid_channels=None):
        super().__init__()
        if not mid_channels:
            mid_channels = out_channels
        self.double_conv = nn.Sequential(
            nn.Conv2d(in_channels, mid_channels, kernel_size=3, padding=1),
            nn.BatchNorm2d(mid_channels),
            nn.ReLU(inplace=True),
            nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True),
        )

    def forward(self, x):
        return self.double_conv(x)


class Down(nn.Module):
    """Downscaling with maxpool then double conv"""

    def __init__(self, in_channels, out_channels):
        super().__init__()
        self.maxpool_conv = nn.Sequential(nn.MaxPool2d(2), DoubleConv(in_channels, out_channels))

    def forward(self, x):
        return self.maxpool_conv(x)


class Up(nn.Module):
    """Upscaling then double conv"""

    def __init__(self, in_channels, out_channels, bilinear=True):
        super().__init__()
        self.up: nn.Upsample | nn.ConvTranspose2d

        # if bilinear, use the normal convolutions to reduce the number of channels
        if bilinear:
            self.up = nn.Upsample(scale_factor=2, mode="bilinear", align_corners=True)
            self.conv = DoubleConv(in_channels, out_channels // 2, in_channels // 2)
        else:
            self.up = nn.ConvTranspose2d(in_channels, in_channels // 2, kernel_size=2, stride=2)
            self.conv = DoubleConv(in_channels, out_channels)

    def forward(self, x1, x2):
        x1 = self.up(x1)
        # input is CHW
        diffY = torch.tensor([x2.size()[2] - x1.size()[2]])
        diffX = torch.tensor([x2.size()[3] - x1.size()[3]])

        x1 = F.pad(
            x1,
            [
                torch.div(diffX, 2, rounding_mode="floor"),
                diffX - torch.div(diffX, 2, rounding_mode="floor"),
                torch.div(diffY, 2, rounding_mode="floor"),
                diffY - torch.div(diffY, 2, rounding_mode="floor"),
            ],
        )
        x = torch.cat([x2, x1], dim=1)
        return self.conv(x)


class OutConv(nn.Module):
    def __init__(self, in_channels, out_channels):
        super().__init__()
        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=1)

    def forward(self, x):
        return self.conv(x)


def convert_pytorch_to_onnx(pytorch_model_path, onnx_model_path):
    # Load the PyTorch model
    model = UNet(n_channels=1, n_classes=1)
    model.load_state_dict(torch.load(pytorch_model_path, map_location="cpu"))
    model.eval()

    # Create dummy input with dynamic size
    dummy_input = torch.randn(1, 1, 512, 512)  # Height [400 to 900], Width fixed at 512

    # Export the model
    torch.onnx.export(
        model,  # model being run
        dummy_input,  # model input (or a tuple for multiple inputs)
        onnx_model_path,  # where to save the model
        export_params=True,  # store the trained parameter weights inside the model file
        opset_version=20,  # the ONNX version to export the model to
        do_constant_folding=True,  # whether to execute constant folding for optimization
        input_names=["input"],  # the model's input names
        output_names=["output"],  # the model's output names
        dynamic_axes={
            "input": {0: "batch_size", 2: "height", 3: "width"},  # variable length axes
            "output": {0: "batch_size", 2: "height", 3: "width"},
        },
    )

    # Verify the model
    onnx_model = onnx.load(onnx_model_path)
    onnx.checker.check_model(onnx_model)

    print(f"Model {pytorch_model_path} converted to {onnx_model_path}")


# List of models to convert
models = [
    "models/case1.pth",
    "models/case2.pth",
]

# Convert each model
for model_path in models:
    onnx_path = model_path.replace(".pth", ".onnx")
    convert_pytorch_to_onnx(model_path, onnx_path)

问题现象

当使用如下dummy输入转换模型时：
```
dummy_input = torch.randn(1, 1, 512, 512)  # Height [400 to 900], Width fixed at 512
```
高度在400-600像素之间的图像推理结果与PyTorch模型一致，但800像素高度的图像推理结果错误，甚至有时无法正常运行。
当使用如下dummy输入转换模型时：
```
dummy_input = torch.randn(1, 1, 885, 512)
```
885像素高度的图像推理结果完全正常。