Specification of one stage emitted by a feature-extracting backbone.
Used by detection / segmentation heads that consume a feature pyramid: each entry describes the channel count and spatial down-sampling factor of one backbone stage relative to the input.
Attributes
stageintZero-based stage index, in the order the backbone emits features
(typically deepening from stage 0 toward higher receptive field).
num_channelsintChannel count of the feature map at this stage.
reductionintSpatial down-sampling factor vs the network input — e.g.
4,
8, 16, 32 for the canonical CNN stage taps.Notes
The layout is timm-compatible: detection heads that consume timm
backbones drop in unchanged. See BackboneMixin for the
protocol that produces a list[FeatureInfo].
Examples
>>> info = FeatureInfo(stage=3, num_channels=512, reduction=32)
>>> info.reduction
32Used by 30
- lucid.models
- lucid.models.vision.alexnet._model
- lucid.models.vision.coatnet._model
- lucid.models.vision.convnext._model
- lucid.models.vision.crossvit._model
- lucid.models.vision.cspnet._model
- lucid.models.vision.cvt._model
- lucid.models.vision.densenet._model
- lucid.models.vision.efficientformer._model
- lucid.models.vision.efficientnet._model
- lucid.models.vision.googlenet._model
- lucid.models.vision.inception._model
… 18 more