TensorRT：PoolLayer的padding

A：Pool层无paddging操作

1：stride=1，不会产生截断不整除问题，所以不用考虑向上或者向下取整问题

2：stride>1，可能或产生截断不整除问题

• 可以使用padding层，补全不能整除的部分

	IActivationLayer* relu1 = basicConv2d(network, weightMap, *scale1->getOutput(0), 64, 7, 2, 3, "conv1.");
	IPaddingLayer* pad1 = network->addPadding(*relu1->getOutput(0), DimsHW{ 0, 0 }, DimsHW{ 1, 1 });
	assert(pad1);
	IPoolingLayer* pool1 = network->addPooling(*pad1->getOutput(0), PoolingType::kMAX, DimsHW{ 3, 3 });
	assert(pool1);
	pool1->setStride(DimsHW{ 2, 2 });

• 可以设置paddingmode，来调整取整的方向问题

enum class PaddingMode : int
{
    kEXPLICIT_ROUND_DOWN = 0, //!< Use explicit padding, rounding output size down.
    kEXPLICIT_ROUND_UP = 1,   //!< Use explicit padding, rounding output size up.
    kSAME_UPPER = 2,          //!< Use SAME padding with prePadding <= postPadding.
    kSAME_LOWER = 3,          /

来源：oschina

链接：https://my.oschina.net/u/4370809/blog/4529166