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

num_classes should be None when box_predictor is specified

Error message

num_classes should be None when box_predictor is specified

What it means

The constructor enforces mutual exclusivity: if num_classes is provided AND a pre-built box_predictor is supplied, it raises 'num_classes should be None when box_predictor is specified' because the predictor already encodes the class count and num_classes would be ambiguous. The complementary branch raises if neither combination is coherent (num_classes None and box_predictor None).

Source

Thrown at pytorch_object_detection/faster_rcnn/network_files/faster_rcnn_framework.py:278

                 box_roi_pool=None, box_head=None, box_predictor=None,
                 # 移除低目标概率      fast rcnn中进行nms处理的阈值   对预测结果根据score排序取前100个目标
                 box_score_thresh=0.05, box_nms_thresh=0.5, box_detections_per_img=100,
                 box_fg_iou_thresh=0.5, box_bg_iou_thresh=0.5,   # fast rcnn计算误差时，采集正负样本设置的阈值
                 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

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. When supplying box_predictor, pass num_classes=None.
2. When you want the framework to build the predictor, pass num_classes and leave box_predictor=None.
3. Standard fine-tune pattern: build model with num_classes, then replace model.roi_heads.box_predictor with your custom predictor afterwards.
4. Check the constructor call for leftover arguments after switching approaches.

Example fix

# before
predictor = FastRCNNPredictor(in_features, num_classes=5)
model = fasterrcnn(backbone=backbone, num_classes=5, box_predictor=predictor)
# after
predictor = FastRCNNPredictor(in_features, num_classes=5)
model = fasterrcnn(backbone=backbone, box_predictor=predictor)  # num_classes omitted (None)

assert not (num_classes is not None and box_predictor is not None), "pass either num_classes or box_predictor, not both"

def predictor_args_ok(num_classes, box_predictor) -> bool:
    return (num_classes is None) == (box_predictor is None)

try:
    model = fasterrcnn(backbone=backbone, num_classes=num_classes, box_predictor=box_predictor)
except ValueError as e:
    if 'num_classes should be None' in str(e):
        model = fasterrcnn(backbone=backbone, box_predictor=box_predictor)
    else:
        raise

When it happens

Trigger: Calling fasterrcnn(backbone=..., num_classes=91, box_predictor=my_predictor) — passing both a custom FastRCNNPredictor and num_classes at once.

Common situations: Fine-tuning workflows where users load a pretrained model, swap in a custom box_predictor for their dataset, but forget to drop num_classes from the constructor call; or copy-pasting both a predictor constructor and a num_classes argument.

Related errors

• In training mode, targets should be passed
• return_layers are not present in model
• backbone should contain an attribute out_channelsspecifying
• num_classes should not be None when box_predictor is not spe
• 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/b897d1da54dfd312. Report an issue: GitHub.