search

chanyn / HKRM

Hybrid Knowledge Routed Module for Large-scale Object Detection (NerIPS2018)
104 stars 19 forks source link

关于代码的一些问题 #12

Open summerrr opened 5 years ago

summerrr commented 5 years ago

TIM图片20190328165208 想问下这个num_A是什么，为什么取10啊，看论文没大看懂 module_A = A_region_compute(5)还有这个5，求解答┭┮﹏┭┮

chanyn commented 5 years ago

num_A is what the number of ensemble adjacent matrixes you want. In our paper, we set num_A = 10 and it is a super-parameter. This idea is inspired by multiple convolution kernels and multi-head. 5 is the dimension of implicit knowledge input. In this paper, we use the output of RPN (dx,dy,dw,dh,score) to capture spatial layout among regions.

okluckychen commented 4 years ago

@summerrr 您好~请问您看懂这个代码了吗？我有很多地方都没有看懂，方便留个联系方式向您请教一下吗？万分感谢！！