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 given AND a prebuilt box_predictor is supplied, the class count is ambiguous (the predictor already encodes its own output size), so it raises this ValueError.

Source

Thrown at pytorch_object_detection/mask_rcnn/network_files/faster_rcnn_framework.py:277

                 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 a custom box_predictor, pass num_classes=None
2. When you want the library to build the predictor, pass box_predictor=None and only num_classes
3. Replace the predictor after construction instead: build the model with num_classes, then swap model.roi_heads.box_predictor

Example fix

// before
model = fasterrcnn_resnet50_fpn(pretrained=True, num_classes=5,
                               box_predictor=FastRCNNPredictor(1024, 5))
// after
model = fasterrcnn_resnet50_fpn(pretrained=True, box_predictor=FastRCNNPredictor(1024, 5))  # 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'

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

When it happens

Trigger: Calling fasterrcnn_resnet50_fpn(pretrained=True, num_classes=..., box_predictor=FastRCNNPredictor(...)) or FasterRCNN(..., num_classes=91, box_predictor=my_predictor) — i.e. specifying both at once.

Common situations: Fine-tuning tutorials where users replace box_predictor but forget to drop num_classes; copy-pasting constructor args from two different customization recipes.

Related errors

• num_classes should not be None when box_predictor is not spe
• illegal stride value.
• The inverted_residual_setting should not be empty.
• The inverted_residual_setting should be List[InvertedResidua
• expected stages_repeats as list of 3 positive ints

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/0437ef70be36078a. Report an issue: GitHub.