AutoFormerV2 类

定义

基于 AutoFormerV2 主干的对象检测网络,仅在 MS COCO 数据集中预先训练。 该网络包含 3 个具有不同参数的缩放:小为 11MB,中型为 21MB,大为 41MB。

public class AutoFormerV2 : TorchSharp.torch.nn.Module<TorchSharp.torch.Tensor,(TorchSharp.torch.Tensor, TorchSharp.torch.Tensor, TorchSharp.torch.Tensor)>
type AutoFormerV2 = class
    inherit torch.nn.Module<torch.Tensor, ValueTuple<torch.Tensor, torch.Tensor, torch.Tensor>>
Public Class AutoFormerV2
Inherits torch.nn.Module(Of torch.Tensor, ValueTuple(Of torch.Tensor, torch.Tensor, torch.Tensor))
继承
torch.nn.Module<TorchSharp.torch.Tensor,ValueTuple<TorchSharp.torch.Tensor,TorchSharp.torch.Tensor,TorchSharp.torch.Tensor>>
AutoFormerV2

构造函数

AutoFormerV2(Int32, List<Int32>, List<Int32>, List<Int32>, torch+Device)

初始化 AutoFormerV2 类的新实例。

方法

forward(torch+Tensor)

基于 AutoFormerV2 主干的对象检测网络,仅在 MS COCO 数据集中预先训练。 该网络包含 3 个具有不同参数的缩放:小为 11MB,中型为 21MB,大为 41MB。

FreezeBN()

冻结网络中 BatchNorm2d 的权重。

适用于