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

illegal stride value.

Error message

illegal stride value.

What it means

InvertedResidual.__init__ (MobileNetV3) validates that the inverted residual config's stride is either 1 or 2 and raises ValueError otherwise. Strides other than 1/2 are not implemented: the block only supports identity/shortcut (stride 1) or strided downsampling with stride 2.

Source

Thrown at pytorch_classification/Test6_mobilenet/model_v3.py:96

        self.expanded_c = self.adjust_channels(expanded_c, width_multi)
        self.out_c = self.adjust_channels(out_c, width_multi)
        self.use_se = use_se
        self.use_hs = activation == "HS"  # whether using h-swish activation
        self.stride = stride

    @staticmethod
    def adjust_channels(channels: int, width_multi: float):
        return _make_divisible(channels * width_multi, 8)


class InvertedResidual(nn.Module):
    def __init__(self,
                 cnf: InvertedResidualConfig,
                 norm_layer: Callable[..., nn.Module]):
        super(InvertedResidual, self).__init__()

        if cnf.stride not in [1, 2]:
            raise ValueError("illegal stride value.")

        self.use_res_connect = (cnf.stride == 1 and cnf.input_c == cnf.out_c)

        layers: List[nn.Module] = []
        activation_layer = nn.Hardswish if cnf.use_hs else nn.ReLU

        # expand
        if cnf.expanded_c != cnf.input_c:
            layers.append(ConvBNActivation(cnf.input_c,
                                           cnf.expanded_c,
                                           kernel_size=1,
                                           norm_layer=norm_layer,
                                           activation_layer=activation_layer))

        # depthwise
        layers.append(ConvBNActivation(cnf.expanded_c,
                                       cnf.expanded_c,
                                       kernel_size=cnf.kernel,

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Set every InvertedResidualConfig stride to 1 or 2 in your configuration list.
2. To downsample more aggressively, insert additional stride-2 blocks instead of using stride > 2.
3. If using the built-in model_name presets ('large'/'small'), don't modify the generated bneck_conf calls.

Example fix

// before
InvertedResidualConfig(16, 3, 24, 24, False, "RE", 3, 1, 1)
// after
InvertedResidualConfig(16, 3, 24, 24, False, "RE", 2, 1, 1)

strides = [cfg.stride for cfg in inverted_residual_setting]
assert all(s in (1, 2) for s in strides), f"illegal strides: {strides}"

def valid_stride(cnf) -> bool:
    return getattr(cnf, 'stride', None) in (1, 2)

try:
    block = InvertedResidual(cnf, norm_layer)
except ValueError as e:
    if "illegal stride" in str(e):
        cnf.stride = 2 if cnf.stride > 1 else 1
        block = InvertedResidual(cnf, norm_layer)
    else:
        raise

When it happens

Trigger: Constructing an InvertedResidualConfig with stride set to 3, 0, or any value outside [1, 2] and then instantiating InvertedResidual(cnf, norm_layer) — typically via a hand-written inverted_residual_setting passed to MobileNetV3.

Common situations: Users customizing the network architecture by editing the bneck configuration list and changing a stride for stronger downsampling; copying configs between MobileNet versions where some blocks allow different strides.

Related errors

• The inverted_residual_setting should not be empty.
• The inverted_residual_setting should be List[InvertedResidua
• illegal stride value.
• expected stages_repeats as list of 3 positive ints
• expected stages_out_channels as list of 5 positive ints

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/48ea0acf9f3f0008. Report an issue: GitHub.