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

illegal stride value.

Error message

illegal stride value.

What it means

In MobileNetV3's InvertedResidual block, the stride from InvertedResidualConfig must be 1 or 2. Any other stride raises this ValueError at construction time. MobileNetV3 inverted residual blocks only support single/double downsampling; larger strides are architecturally undefined here.

Source

Thrown at pytorch_segmentation/lraspp/src/mobilenet_backbone.py:101

        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
        self.dilation = dilation

    @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
        stride = 1 if cnf.dilation > 1 else cnf.stride
        layers.append(ConvBNActivation(cnf.expanded_c,
                                       cnf.expanded_c,

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Set each config's stride to 1 or 2 only
2. If you need more downsampling, add more blocks or adjust input image stride via the first ConvBNActivation layer instead
3. Re-validate any generated/serialized InvertedResidualConfig list before constructing MobileNetV3

Example fix

// before
cnf = InvertedResidualConfig(16, 3, 64, 64, True, 'RE', 4, 1, 1)  # stride=4
// after
cnf = InvertedResidualConfig(16, 3, 64, 64, True, 'RE', 2, 1, 1)  # stride in {1,2}

for cnf in inverted_residual_setting:
    assert cnf.stride in (1, 2), f"bad stride {cnf.stride} in {cnf}"

def strides_valid(cfgs):
    return all(getattr(c, 'stride', None) in (1, 2) for c in cfgs)

try:
    model = MobileNetV3(inverted_residual_setting=cfgs, num_classes=nc)
except ValueError as e:
    logging.error("invalid block config: %s", e)
    raise SystemExit(1)

When it happens

Trigger: Hand-building `InvertedResidualConfig(cnf)` with `stride=3` (or 0, 4) and passing to `MobileNetV3(inverted_residual_setting=[cnf,...])`; editing the predefined bneck_cfg list for a custom resolution; loading a config from JSON where stride was wrongly specified.

Common situations: Custom backbone tuning for segmentation (developers try stride=4 to downsample faster); mis-ordered positional args when constructing InvertedResidualConfig manually (width/multiplier vs stride confusion); auto-generated NAS-style configs with unsupported strides.

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

• The inverted_residual_setting should not be empty.
• illegal stride value.
• return_layers are not present in model
• return_layers are not present in model
• The inverted_residual_setting should be List[InvertedResidua

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/a4d03ef5914ce7f5. Report an issue: GitHub.