论文中所说的BN层

[Gao-JT]

作者你好！在SiamVGG论文中Section 4.2中提到“we notice the other one (without batch normalization) achieves much higher performance in tracking tasks with 0.654 AUC of success plots for OPE than the model with batch normalization (with 0.589 AUC).” 这里所说的BN层是指在VGG模型中每个ConV层后面的BN层，还是指最后的bn_adjust层（BN层）？

[Gao-JT]

已经通过代码实验确认了，感谢作者！

[sunzhen6251]

@Gao-JT 你好，请问你能复现该模型的性能么？我在python3.7和pytorch 1.4上可以跑起来该模型，但效果不佳。