Microsoft.ML.TorchSharp.AutoFormerV2 Namespace

Classes

Anchors

Anchor boxes are a set of predefined bounding boxes of a certain height and width, whose location and size can be adjusted by the regression head of model.

Attention

The Attention layer.

AutoFormerV2

The object detection network based on AutoFormerV2 backbone which is pretrained only in MS COCO dataset. The network contains 3 scales with different parameters: small for 11MB, medium for 21MB and large for 41MB.

AutoFormerV2Backbone

The backbone of AutoFormerV2 object detection network.

AutoFormerV2Block

The block module of AutoFormer network.

BasicLayer

The basic layer.

Conv2dBN

The Convolution and BN module.

ConvLayer

The layer of convolution blocks in AutoFormer network.

ConvModule

The convolution and activation module.

FPN

The FPN (Feature Pyramid Networks) layer.

MBConv

The MBConv layer.

MLP

The MLP (Multilayer Perceptron) layer.

ObjectDetectionMetrics
ObjectDetectionTrainer
ObjectDetectionTrainer.Options
ObjectDetectionTransformer
PatchEmbed

The PatchEmbed layer.

PatchMerging

The PatchMerging layer.

RetinaHead

The head of RetinaNet.