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

num_classes should not be None when box_predictor is not spe

Error message

num_classes should not be None when box_predictor is not specified

What it means

FasterRCNN.__init__ requires exactly one of two configuration modes: either build the predictor internally from `num_classes`, or receive a fully-built `box_predictor`. Passing neither (num_classes=None and box_predictor=None) is ambiguous, so the constructor raises ValueError to fail fast.

Source

Thrown at pytorch_object_detection/faster_rcnn/network_files/faster_rcnn_framework.py:282

                 box_batch_size_per_image=512, box_positive_fraction=0.25,  # fast rcnn计算误差时采样的样本数，以及正样本占所有样本的比例
                 bbox_reg_weights=None):
        if not hasattr(backbone, "out_channels"):
            raise ValueError(
                "backbone should contain an attribute out_channels"
                "specifying the number of output channels  (assumed to be the"
                "same for all the levels"
            )

        assert isinstance(rpn_anchor_generator, (AnchorsGenerator, type(None)))
        assert isinstance(box_roi_pool, (MultiScaleRoIAlign, type(None)))

        if num_classes is not None:
            if box_predictor is not None:
                raise ValueError("num_classes should be None when box_predictor "
                                 "is specified")
        else:
            if box_predictor is None:
                raise ValueError("num_classes should not be None when box_predictor "
                                 "is not specified")

        # 预测特征层的channels
        out_channels = backbone.out_channels

        # 若anchor生成器为空，则自动生成针对resnet50_fpn的anchor生成器
        if rpn_anchor_generator is None:
            anchor_sizes = ((32,), (64,), (128,), (256,), (512,))
            aspect_ratios = ((0.5, 1.0, 2.0),) * len(anchor_sizes)
            rpn_anchor_generator = AnchorsGenerator(
                anchor_sizes, aspect_ratios
            )

        # 生成RPN通过滑动窗口预测网络部分
        if rpn_head is None:
            rpn_head = RPNHead(
                out_channels, rpn_anchor_generator.num_anchors_per_location()[0]
            )

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Pass num_classes=N (including background class, e.g. 91 for COCO, 21 for VOC) when constructing the network from a backbone
2. Or pass a pre-built box_predictor (e.g. taken from the pretrained fast_rcnn_predictor) instead of num_classes
3. Ensure the backbone passed in exposes an `out_channels` attribute as required by the constructor

Example fix

// before
model = FasterRCNN(backbone)
// after
model = FasterRCNN(backbone, num_classes=91)
// or
model = FasterRCNN(backbone, box_rpn=None, box_predictor=pretrained_predictor)

assert (num_classes is not None) != (box_predictor is not None), "Provide exactly one of num_classes or box_predictor"
model = FasterRCNN(backbone, num_classes=num_classes, box_predictor=box_predictor)

def has_predictor_config(model_cfg) -> bool:
    return (model_cfg.get("num_classes") is not None) != (model_cfg.get("box_predictor") is not None)

try:
    model = FasterRCNN(backbone, num_classes=num_classes)
except ValueError as e:
    if "num_classes should not be None" in str(e):
        model = FasterRCNN(backbone, num_classes=num_classes or 91)

When it happens

Trigger: Calling FasterRCNN(backbone, ...) with neither num_classes nor box_predictor supplied, e.g. FasterRCNN(backbone) after copying only the backbone from a pretrained model.

Common situations: Copying a backbone from a pretrained model and forgetting to pass either the pretrained box_predictor or the new num_classes; refactoring code that removed one argument but not the other.

Related errors

• expected stages_repeats as list of 3 positive ints
• expected stages_out_channels as list of 5 positive ints
• image: {} isn't RGB mode.
• dataset have {} classes, but input {}
• dataset have {} classes, but input {}

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/8a2d8acd068597e2. Report an issue: GitHub.