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

target should not be None.

Error message

target should not be None.

What it means

During training, ROIHeads.select_training_samples needs ground-truth targets to match proposals against and compute classification/box losses. The code checks targets via check_targets first, then raises ValueError if the targets argument itself is None, since training cannot proceed without annotations.

Source

Thrown at pytorch_object_detection/mask_rcnn/network_files/roi_head.py:322

                                proposals,  # type: List[Tensor]
                                targets     # type: Optional[List[Dict[str, Tensor]]]
                                ):
        # type: (...) -> Tuple[List[Tensor], List[Tensor], List[Tensor], List[Tensor]]
        """
        划分正负样本，统计对应gt的标签以及边界框回归信息
        list元素个数为batch_size
        Args:
            proposals: rpn预测的boxes
            targets:

        Returns:

        """

        # 检查target数据是否为空
        self.check_targets(targets)
        if targets is None:
            raise ValueError("target should not be None.")

        dtype = proposals[0].dtype
        device = proposals[0].device

        # 获取标注好的boxes以及labels信息
        gt_boxes = [t["boxes"].to(dtype) for t in targets]
        gt_labels = [t["labels"] for t in targets]

        # append ground-truth bboxes to proposal
        # 将gt_boxes拼接到proposal后面
        proposals = self.add_gt_proposals(proposals, gt_boxes)

        # get matching gt indices for each proposal
        # 为每个proposal匹配对应的gt_box，并划分到正负样本中
        matched_idxs, labels = self.assign_targets_to_proposals(proposals, gt_boxes, gt_labels)
        # sample a fixed proportion of positive-negative proposals
        # 按给定数量和比例采样正负样本
        sampled_inds = self.subsample(labels)

View on GitHub (pinned to 1ec3fe6f37)

Solutions

1. Pass a non-None list of target dicts (with 'boxes' and 'labels') in training mode
2. Verify the dataloader actually yields targets and filters out empty samples
3. Only omit targets during inference under torch.no_grad() with model.eval()

Example fix

// before
loss_dict = model(images)
// after
loss_dict = model(images, targets)  # targets = [{'boxes': ..., 'labels': ..., 'masks': ...}]

if model.training:
    assert targets is not None and all('boxes' in t and 'labels' in t for t in targets), "targets required in train mode"
    losses = model(images, targets)

def has_valid_targets(targets):
    return targets is not None and isinstance(targets, list) and all(isinstance(t, dict) and 'boxes' in t for t in targets)

try:
    losses = model(images, targets)
except ValueError as e:
    if 'target' in str(e): targets = load_targets(batch)
    raise

When it happens

Trigger: Calling model(images) in train() mode (model.train()) without passing targets, e.g. model(images) instead of model(images, targets).

Common situations: Forgetting to pass targets in the training loop; a dataloader returning None targets for unlabeled samples; reusing inference code for training.

Understand the failure class

Background: "missing required argument" and "the following required arguments were not provided": what required-argument errors mean and how to fix them — this error's family across 20 libraries.

Related errors

• if in training, matched_idxs should not be None
• targets, pos_matched_idxs, mask_logits cannot be None when t
• num_classes should be None when mask_predictor is specified
• expected stages_repeats as list of 3 positive ints
• expected stages_out_channels as list of 5 positive ints

AI-assisted analysis of WZMIAOMIAO/deep-learning-for-image-processing@1ec3fe6f37 (2026-08-30). Data as JSON: /api/errors/43dcc24666f4c4b5. Report an issue: GitHub.