lllyasviel/Fooocus · error · NotImplementedError

network_name={network_name}

Error message

network_name={network_name}

What it means

Raised by facexlib's generate_config() when building a RetinaFace detection model with a network_name that is neither 'mobile0.25' nor 'resnet50'. The factory only ships two backbone configs, so any other string falls through to NotImplementedError. It fires inside RetinaFace.__init__ (and init_detection_model), i.e. at model construction time before any inference.

Source

Thrown at extras/facexlib/detection/retinaface.py:68

        'epoch': 100,
        'decay1': 70,
        'decay2': 90,
        'image_size': 840,
        'return_layers': {
            'layer2': 1,
            'layer3': 2,
            'layer4': 3
        },
        'in_channel': 256,
        'out_channel': 256
    }

    if network_name == 'mobile0.25':
        return cfg_mnet
    elif network_name == 'resnet50':
        return cfg_re50
    else:
        raise NotImplementedError(f'network_name={network_name}')


class RetinaFace(nn.Module):

    def __init__(self, network_name='resnet50', half=False, phase='test', device=None):
        self.device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') if device is None else device

        super(RetinaFace, self).__init__()
        self.half_inference = half
        cfg = generate_config(network_name)
        self.backbone = cfg['name']

        self.model_name = f'retinaface_{network_name}'
        self.cfg = cfg
        self.phase = phase
        self.target_size, self.max_size = 1600, 2150
        self.resize, self.scale, self.scale1 = 1., None, None
        self.mean_tensor = torch.tensor([[[[104.]], [[117.]], [[123.]]]], device=self.device)

View on GitHub (pinned to ae05379cc9)

Solutions

1. Use exactly 'mobile0.25' (lightweight) or 'resnet50' (accurate) as network_name.
2. Check for typos/case differences in the config value passed to init_detection_model or RetinaFace.
3. If you truly need another backbone, add a cfg dict for it in extras/facexlib/detection/retinaface.py and return it from generate_config(), plus a matching pretrained URL.
4. Downstream (e.g. GFPGAN/CodeFormer restore) usually requires 'retinaface_resnet50'; pass that name.

Example fix

// before
model = init_detection_model('resnet18', device=device)

// after
model = init_detection_model('resnet50', device=device)  # or 'mobile0.25'

VALID_NETWORKS = {'mobile0.25', 'resnet50'}
if network_name not in VALID_NETWORKS:
    raise ValueError(f'network_name must be one of {sorted(VALID_NETWORKS)}, got {network_name!r}')

def is_valid_retinaface_network(name: str) -> bool:
    return name in {'mobile0.25', 'resnet50'}

try:
    model = init_detection_model(network_name, device=device)
except NotImplementedError as e:
    logger.error('Unsupported RetinaFace backbone: %s (use mobile0.25 or resnet50)', network_name)
    raise

When it happens

Trigger: Calling RetinaFace(network_name='resnet18') / init_detection_model(det_model='resnet18', ...) or any typo like 'MobileNet0.25', 'mobile0_25', 'resnet101'. Only exact strings 'mobile0.25' and 'resnet50' are accepted.

Common situations: Copying a model name from another repo (e.g. insightface's det names or IR-SE variants), typo in YAML/CLI config, or trying to plug a custom backbone into facexlib without extending generate_config.

Related errors

• {model_name} is not implemented.
• {model_name} is not implemented.
• Max depth of recursive function `tie_encoder_to_decoder` rea
• No paddings to do, output_size must be None or {}
• Not (0 <= inner_padding_factor <= 1.0)

AI-assisted analysis of lllyasviel/Fooocus@ae05379cc9 (2026-08-15). Data as JSON: /api/errors/c22d78e3d5c1249d. Report an issue: GitHub.