Number1Tao

use im2col_step in DeformableConvolutionOp, so we wonder : why not use im2col_step in ModulatedDeformableConvolutionOp?

A bug？

1

act1 actually not defined 

which caffe you use? Official caffe scale layer do NOT contain batch normalize

fix bug:Incorrect file directory

1

hi，when I run your code ,i find a small bug: File "/usr/local/lib/python2.7/dist-packages/tensorflow/python/framework/errors.py", line 450, in raise_exception_on_not_ok_status pywrap_tensorflow.TF_GetCode(status)) tensorflow.python.framework.errors.NotFoundError: saved_networks/saved_networks This is because the file path “tetris-dqn-31660000” is wrong，after delete path"saved_networks",bug...

Hi, As the complexity of the model increases, the dice coefficient decreases using more datasets (23 datasets) pretrained model.  However using other datasets (8 datasets?) pretrained model, as the...

正排索引维护了文档ID和文档中出现的词项之间的一一映射关系，也就是说一篇文档中出现了哪些词项以及各个词项在文档中出现的词频tf均维护在正排索引表中，表结构用HashMap进行维护，词项集合用ArrayList进行维护。值得注意的是为了降低词项字符串重复的内存开销，在此用HashSet新维护了一个字典表，索引为该词项，正排表以及后续的倒排表中所有出现过的词项均是对该字典表中的词项的一份引用。故大大降低了词项重复所导致的内存空间占用，经测试内存开销降低了约30%。