# backbones

Python API: `luxonis_train.nodes.backbones`

将图像转换为特征图的骨干网络。

大多数骨干网络返回特征图列表。步幅从一个图到下一个图增加。颈部或头部随后读取这些图。一些骨干网络有所不同：

 * [ContextSpatial](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/contextspatial.md)
   和
   [DDRNet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/ddrnet/ddrnet.md)
   返回输入大小 1/8 的特征，用于分割头部。
 * [DinoV3](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/dinov3/dinov3.md)
   在 `return_sequence` 为 `True` 时返回一个包含一个 CLS 嵌入的列表，而不是特征图。
 * [GhostFaceNet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/ghostfacenet/ghostfacenet.md)
   目标是人脸嵌入。
 * [RecSubNet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/recsubnet/recsubnet.md)
   返回一个包含图像重建和原始输入的包，用于异常检测。

具有变体的骨干网络使用 `variant` 来选择大小。每个骨干网络的文档字符串列出了其变体以及每个变体设置的参数。

## 子页面

 * [ddrnet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/ddrnet.md)
 * [dinov3](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/dinov3.md)
 * [efficientrep](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/efficientrep.md)
 * [efficientvit](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/efficientvit.md)
 * [ghostfacenet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/ghostfacenet.md)
 * [micronet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/micronet.md)
 * [mobileone](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/mobileone.md)
 * [pplcnet_v3](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/pplcnet_v3.md)
 * [recsubnet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/recsubnet.md)
 * [repvgg](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/repvgg.md)
 * [contextspatial](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/contextspatial.md)
 * [efficientnet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/efficientnet.md)
 * [mobilenetv2](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/mobilenetv2.md)
 * [resnet](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/resnet.md)
 * [rexnetv1](/software-v3/ai-inference/model-source/training/luxonis-train/luxonis-train-api-reference/nodes/backbones/rexnetv1.md)
