It's because the pytorch implementation of inception v3 uses an adaptive average pooling layer right before the fully-connected layer.
確定! 回上一頁