open-mmlab/mmdetection · warning

``ignore_label`` and ``loss_weight`` would be deprecated soo

Error message

``ignore_label`` and ``loss_weight`` would be deprecated soon. Please set ``ingore_index`` and ``loss_weight`` in ``loss_seg`` instead.

What it means

Warning from FusedSemanticHead.__init__: the standalone `ignore_label` and `loss_weight` constructor arguments are being deprecated; they are merely copied into loss_seg['ignore_index'] / loss_seg['loss_weight']. Set them inside the loss_seg dict instead.

Source

Thrown at mmdet/models/roi_heads/mask_heads/fused_semantic_head.py:97

                    in_channels,
                    conv_out_channels,
                    3,
                    padding=1,
                    conv_cfg=self.conv_cfg,
                    norm_cfg=self.norm_cfg))
        self.conv_embedding = ConvModule(
            conv_out_channels,
            conv_out_channels,
            1,
            conv_cfg=self.conv_cfg,
            norm_cfg=self.norm_cfg)
        self.conv_logits = nn.Conv2d(conv_out_channels, self.num_classes, 1)
        if ignore_label:
            loss_seg['ignore_index'] = ignore_label
        if loss_weight:
            loss_seg['loss_weight'] = loss_weight
        if ignore_label or loss_weight:
            warnings.warn('``ignore_label`` and ``loss_weight`` would be '
                          'deprecated soon. Please set ``ingore_index`` and '
                          '``loss_weight`` in ``loss_seg`` instead.')
        self.criterion = MODELS.build(loss_seg)

    def forward(self, feats: Tuple[Tensor]) -> Tuple[Tensor]:
        """Forward function.

        Args:
            feats (tuple[Tensor]): Multi scale feature maps.

        Returns:
            tuple[Tensor]:

                - mask_preds (Tensor): Predicted mask logits.
                - x (Tensor): Fused feature.
        """
        x = self.lateral_convs[self.fusion_level](feats[self.fusion_level])
        fused_size = tuple(x.shape[-2:])

View on GitHub (pinned to cfd5d3a985)

Solutions

1. Move ignore_label into loss_seg as ignore_index (e.g. loss_seg=dict(type='CrossEntropyLoss', ignore_index=-1))
2. Move loss_weight into loss_seg as loss_weight
3. Remove the top-level ignore_label/loss_weight keys from the head config

Example fix

# before
roi_head=dict(..., semantic_head=dict(type='FusedSemanticHead', ignore_label=-1, loss_weight=0.2))
# after
roi_head=dict(..., semantic_head=dict(type='FusedSemanticHead', loss_seg=dict(type='CrossEntropyLoss', ignore_index=-1, loss_weight=0.2)))

assert 'ignore_label' not in sem_head_cfg and 'loss_weight' not in sem_head_cfg, 'set ignore_index/loss_weight inside loss_seg'

When it happens

Trigger: Constructing FusedSemanticHead with ignore_label=-1 (typical for Panoptic FPN / HTC configs) and/or an explicit loss_weight argument; any config inherited from older mask-rcnn/htc configs.

Common situations: Upgrading configs from older mmdet where ignore_label/loss_weight were top-level head args; panoptic segmentation setups.

Related errors

• pretrained must be a str or None
• DeprecationWarning: pretrained is deprecated, please use "in
• DeprecationWarning: `num_anchors` is deprecated, for consist
• DeprecationWarning: anchor_generator is deprecated, please u
• You are calling `simple_test_bboxes` in `dense_test_mixins`,

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/3205b3dc4b7e4e7f. Report an issue: GitHub.