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

backbone is None

Error message

backbone is None

What it means

SSD300.__init__ raises Exception('backbone is None') when instantiated without a backbone argument. SSD300 extracts features from a pretrained backbone (e.g. resnet50-fpn-style) and has no default one; the constructor validates it first.

Source

Thrown at pytorch_object_detection/ssd/src/ssd_model.py:36

        self.feature_extractor = nn.Sequential(*list(net.children())[:7])

        conv4_block1 = self.feature_extractor[-1][0]

        # 修改conv4_block1的步距，从2->1
        conv4_block1.conv1.stride = (1, 1)
        conv4_block1.conv2.stride = (1, 1)
        conv4_block1.downsample[0].stride = (1, 1)

    def forward(self, x):
        x = self.feature_extractor(x)
        return x


class SSD300(nn.Module):
    def __init__(self, backbone=None, num_classes=21):
        super(SSD300, self).__init__()
        if backbone is None:
            raise Exception("backbone is None")
        if not hasattr(backbone, "out_channels"):
            raise Exception("the backbone not has attribute: out_channel")
        self.feature_extractor = backbone

        self.num_classes = num_classes
        # out_channels = [1024, 512, 512, 256, 256, 256] for resnet50
        self._build_additional_features(self.feature_extractor.out_channels)
        self.num_defaults = [4, 6, 6, 6, 4, 4]
        location_extractors = []
        confidence_extractors = []

        # out_channels = [1024, 512, 512, 256, 256, 256] for resnet50
        for nd, oc in zip(self.num_defaults, self.feature_extractor.out_channels):
            # nd is number_default_boxes, oc is output_channel
            location_extractors.append(nn.Conv2d(oc, nd * 4, kernel_size=3, padding=1))
            confidence_extractors.append(nn.Conv2d(oc, nd * self.num_classes, kernel_size=3, padding=1))

        self.loc = nn.ModuleList(location_extractors)

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Pass a backbone: SSD300(backbone=resnet50_fpn_backbone(), num_classes=...) Check torchvision version for resnet50_fpn_backbone availability (moved to torchvision.models.detection.backbone_utils in newer versions) Add an assertion before construction that the backbone factory returned non-None

Example fix

// before
model = SSD300(num_classes=21)
// after
backbone = resnet50_fpn_backbone(pretrain_path='resnet50.pth')
model = SSD300(backbone=backbone, num_classes=21)

backbone = resnet50_fpn_backbone()
assert backbone is not None, 'backbone factory returned None'
model = SSD300(backbone=backbone, num_classes=21)

def build_ssd(backbone, num_classes):
    assert backbone is not None, 'SSD300 requires a backbone'
    from pytorch_object_detection.ssd.src.ssd_model import SSD300
    return SSD300(backbone=backbone, num_classes=num_classes)

try:
    model = SSD300(backbone=backbone, num_classes=21)
except Exception as e:
    print(f'Model construction failed: {e}; ensure backbone passed')

When it happens

Trigger: Calling SSD300() or SSD300(num_classes=21) with no backbone kwarg; a factory function returning None backbone due to a failed pretrained-weight load path.

Common situations: Following a tutorial snippet that shows SSD300(num_classes=...) only; torchvision backbone builder behind a version-dependent import failing silently and returning None; forgetting to pass create_backbone() result.

Related errors

• the backbone not has attribute: out_channel
• return_layers are not present in model
• backbone should contain an attribute out_channelsspecifying
• backbone should contain an attribute out_channels specifying
• In training mode, targets should be passed

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/5e0b424d838aee09. Report an issue: GitHub.