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

illegal stride value.

Error message

illegal stride value.

What it means

InvertedResidual (EfficientNet MBConv block) only supports strides 1 and 2 because shortcut connection logic (use_res_connect) and the downsampling conv layers are designed for those values. Any other stride in the InvertedResidualConfig is rejected in __init__ with a ValueError at model-construction time.

Source

Thrown at pytorch_classification/Test9_efficientNet/model.py:141

        self.out_c = self.adjust_channels(out_c, width_coefficient)
        self.use_se = use_se
        self.stride = stride
        self.drop_rate = drop_rate
        self.index = index

    @staticmethod
    def adjust_channels(channels: int, width_coefficient: float):
        return _make_divisible(channels * width_coefficient, 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 = OrderedDict()
        activation_layer = nn.SiLU  # alias Swish

        # expand
        if cnf.expanded_c != cnf.input_c:
            layers.update({"expand_conv": ConvBNActivation(cnf.input_c,
                                                           cnf.expanded_c,
                                                           kernel_size=1,
                                                           norm_layer=norm_layer,
                                                           activation_layer=activation_layer)})

        # depthwise
        layers.update({"dwconv": 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 (use stride 2 only on the first block of each stage for downsampling)
2. Remove or replace the custom block with a different module if you need stride > 2, e.g. stack two stride-2 blocks or use pooling
3. Validate the config table values before constructing the model

Example fix

// before
InvertedResidualConfig(input_c, kernel=3, expanded_c, out_c, use_se=True, activation='silu', stride=3)
// after
InvertedResidualConfig(input_c, kernel=3, expanded_c, out_c, use_se=True, activation='silu', stride=2)

for cnf in inverted_residual_setting:
    if cnf.stride not in (1, 2):
        raise ValueError(f'stride must be 1 or 2, got {cnf.stride}')

def valid_stride(cnf) -> bool:
    return cnf.stride in (1, 2)

try:
    model = efficientnet(num_classes=5)
except ValueError as e:
    if 'illegal stride' in str(e):
        print('Fix the stride in your block config table:', e)
    raise

When it happens

Trigger: Building an EfficientNet variant whose _make_layers / configuration table supplies cnf.stride of 3, 4, 0, or a float instead of 1 or 2 when instantiating InvertedResidual.

Common situations: Hand-editing the efficientnet_config list to scale the network; porting block definitions from another architecture (e.g. HRNet strided blocks); typo like stride=22 instead of 2.

Related errors

• illegal stride value.
• expected stages_repeats as list of 3 positive ints
• expected stages_out_channels as list of 5 positive ints
• not support model name: {}
• image: {} isn't RGB mode.

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/2d66219183788d81. Report an issue: GitHub.