PaddlePaddle/PaddleOCR · error · NotImplementedError

mode[{mode}_model] is not implemented!

Error message

mode[{mode}_model] is not implemented!

What it means

MicroNet recognition backbone implements four fixed architectures selected by the mode string: 'M0', 'M1', 'M2', 'M3' (the SOURCE shows the M2/M3 branches setting input_channel, stem_groups, out_ch, activation_cfg). Any other value raises NotImplementedError('mode[<mode>_model] is not implemented!') at construction.

Source

Thrown at ppocr/modeling/backbones/rec_micronet.py:551

            input_channel = 6
            stem_groups = 3, 2
            out_ch = 576
            activation_cfg["init_a"] = 1.0, 1.0
            activation_cfg["init_b"] = 0.0, 0.0
        elif mode == "M2":
            input_channel = 8
            stem_groups = 4, 2
            out_ch = 768
            activation_cfg["init_a"] = 1.0, 1.0
            activation_cfg["init_b"] = 0.0, 0.0
        elif mode == "M3":
            input_channel = 12
            stem_groups = 4, 3
            out_ch = 432
            activation_cfg["init_a"] = 1.0, 0.5
            activation_cfg["init_b"] = 0.0, 0.5
        else:
            raise NotImplementedError("mode[" + mode + "_model] is not implemented!")

        layers = [StemLayer(3, input_channel, stride=2, groups=stem_groups)]

        for idx, val in enumerate(self.cfgs):
            s, n, c, ks, c1, c2, g1, g2, c3, g3, g4, y1, y2, y3, r = val

            t1 = (c1, c2)
            gs1 = (g1, g2)
            gs2 = (c3, g3, g4)
            activation_cfg["dy"] = [y1, y2, y3]
            activation_cfg["ratio"] = r

            output_channel = c
            layers.append(
                DYMicroBlock(
                    input_channel,
                    output_channel,
                    kernel_size=ks,

View on GitHub (pinned to 2661c7c0ef)

Solutions

1. Set mode to exactly one of 'M0', 'M1', 'M2', 'M3' (uppercase M) in the rec_micronet backbone config
2. Remove MobileNetV3-only keys (model_name, scale) from the MicroNet backbone block
3. Prefer M2/M3 as in the shipped configs; verify dims also match a supported preset

Example fix

# before
Backbone:
  name: MicroNet
  mode: m1        # lowercase -> NotImplementedError

# after
Backbone:
  name: MicroNet
  mode: M1

assert mode in ('M0', 'M1', 'M2', 'M3'), f'MicroNet mode must be M0-M3, got {mode!r}'

def is_micronet_mode(m: str) -> bool:
    return m in {'M0', 'M1', 'M2', 'M3'}

When it happens

Trigger: Backbone config with mode='m0' (lowercase), mode='M4', or a typo like 'MO'; also passing the broader model_name/scale-style keys used by MobileNetV3 configs, which MicroNet does not accept.

Common situations: Copy-pasting a MobileNetV3 backbone block and only changing name: MicroNet, leaving incompatible keys; case-sensitive YAML values; assuming more variants exist than the four implemented.

Related errors

• The mixer must be one of [Global, Local, Conv]
• {} is not supported in MultiLoss yet
• mode[{model_name}_model] is not implemented!
• RecResizeImg.image_shape is required in rec inference.yml
• Unexpected recognition channels: ${String(channels)}

AI-assisted analysis of PaddlePaddle/PaddleOCR@2661c7c0ef (2026-08-14). Data as JSON: /api/errors/e2015f53a4b7304a. Report an issue: GitHub.