（beta）torch_npu.contrib.DCNv2

接口原型

torch_npu.contrib.DCNv2(in_channels, out_channels, kernel_size, stride=1, padding=0, dilation=1, groups=1, deformable_groups=1, bias=True, pack=True)

功能描述

应用基于NPU的调制可变形2D卷积操作。ModulationDeformConv的实现主要是基于mmcv的实现进行设计和重构。

参数说明

• in_channels (int)：Number of channels in the input image.
• out_channels (int)：Number of channels produced by the convolution.
• kernel_size(int, tuple)：Size of the convolving kernel.
• stride(int, tuple)：Stride of the convolution. Default: 1.
• padding (int or tuple)：Zero-padding added to both sides of the input.
• Default：0.
• dilation (int or tuple)：Spacing between kernel elements. Default：1.
• groups (int)：Number of blocked connections from input.
• channels to output channels. Default：1.
• deform_groups (int)：Number of deformable group partitions.
• bias (bool)：If True, adds a learnable bias to the output. Default：False.
• pack (bool)：If True, conv_offset and mask will be included in this module. Default：True.

约束说明

ModedDeformConv仅实现fp32数据类型下的操作。注意，con_offset中的权重和偏置必须初始化为0。

支持的型号

• Atlas 训练系列产品
• Atlas A2 训练系列产品
• Atlas 推理系列产品

调用示例

>>> x = torch.randn((2, 2, 5, 5), dtype=torch.float32).npu()
>>> x.requires_grad = True
>>> model = torch_npu.contrib.DCNv2(2, 2, 3, 2, 1).npu()
>>> output = model(x)
>>> output.sum().backward()