WZMIAOMIAO/deep-learning-for-image-processing · error · ValueError

replace_stride_with_dilation should be None or a 3-element t

Error message

replace_stride_with_dilation should be None or a 3-element tuple, got {}

What it means

ResNet backbone construction validates the `replace_stride_with_dilation` argument. It may be None (defaults to [False, False, False]) or a sequence of exactly 3 booleans — one per layer2/3/4 stride block. Passing anything whose length is not 3 raises this ValueError immediately in `_make_layer` setup of ResNet.__init__.

Source

Thrown at pytorch_segmentation/fcn/src/backbone.py:82

class ResNet(nn.Module):

    def __init__(self, block, layers, num_classes=1000, zero_init_residual=False,
                 groups=1, width_per_group=64, replace_stride_with_dilation=None,
                 norm_layer=None):
        super(ResNet, self).__init__()
        if norm_layer is None:
            norm_layer = nn.BatchNorm2d
        self._norm_layer = norm_layer

        self.inplanes = 64
        self.dilation = 1
        if replace_stride_with_dilation is None:
            # each element in the tuple indicates if we should replace
            # the 2x2 stride with a dilated convolution instead
            replace_stride_with_dilation = [False, False, False]
        if len(replace_stride_with_dilation) != 3:
            raise ValueError("replace_stride_with_dilation should be None "
                             "or a 3-element tuple, got {}".format(replace_stride_with_dilation))
        self.groups = groups
        self.base_width = width_per_group
        self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3,
                               bias=False)
        self.bn1 = norm_layer(self.inplanes)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
                                       dilate=replace_stride_with_dilation[0])
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
                                       dilate=replace_stride_with_dilation[1])
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
                                       dilate=replace_stride_with_dilation[2])
        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(512 * block.expansion, num_classes)

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Pass exactly a 3-element tuple/list of booleans, e.g. (False, True, True)
2. Or pass None to use default strides everywhere
3. Verify each boolean maps to layer2/layer3/layer4 stride replacement in order

Example fix

// before
backbone = resnet50_fpn_backbone(replace_stride_with_dilation=[True, True])
// after
backbone = resnet50_fpn_backbone(replace_stride_with_dilation=(False, True, True))

def check_dilation(replace_stride_with_dilation):
    assert replace_stride_with_dilation is None or (
        len(replace_stride_with_dilation) == 3 and all(isinstance(b, bool) for b in replace_stride_with_dilation))
check_dilation(my_cfg)

def is_valid_dilation(v):
    return v is None or (isinstance(v, (list, tuple)) and len(v) == 3 and all(isinstance(b, bool) for b in v))

try:
    backbone = resnet50_fpn_backbone(replace_stride_with_dilation=cfg.dilation)
except ValueError as e:
    logging.error("bad dilation config %s: %s", cfg.dilation, e)
    backbone = resnet50_fpn_backbone()

When it happens

Trigger: Passing `replace_stride_with_dilation=[True, True]` or `[True]`; passing a tuple with more than 3 elements; passing an empty list; copying torchvision code that used a different number of stages.

Common situations: Building dilated FCN/DeepLab backbones where developers enable dilation on some stages but forget the third element; typos when copying torchvision ResNet examples; configuring a 2-stage custom ResNet variant but reusing 3-element validation.

Understand the failure class

Background: "Must be a positive integer", "Invalid value", "Unsupported": the invalid-argument-value error family, when a library rejects the value you pass — this error's family across 35 libraries.

Related errors

• replace_stride_with_dilation should be None or a 3-element t
• the cfg file not exist...
• Unsupported fields:{} in cfg
• conv2d filter size must be int type.
• return_layers are not present in model

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/e03a05678ebf58fa. Report an issue: GitHub.