nihui
/
ncnn
Not watched
1
0
Fork 0
Code
Releases 46 Wiki evaluate Activity
Issues 0 Pull Requests 0 Datasets Model Cloudbrain HPC
247 Commits
5 Branches
354 MB
2 Download
Tree: db5e805eff
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'db5e805eff'
${ noResults }
Commit Graph

158 Commits (db5e805effe6df39462ade72c2c438331a25a0c8)

Author SHA1 Message Date
  nihuini db5e805eff padding_mode for Pooling, fix #261 8 years ago
  nihui 2d9410742b concat slice shufflechannel honor elemsize 8 years ago
  nihui 8ccae1d4fd prevent reuse of param array, fix #258 8 years ago
  nihuini 75218953cc aarch64 assembly for conv1x1s1, unroll outch inch as 8x8 8 years ago
  nihuini 76a55693a6 decouple convolutiondepthwise and convolution, reduce binary size by 10%, fix #254 8 years ago
  nihuini 3ffb502bc6 reuse if the same shape 8 years ago
  nihuini c6506d6ecd remaining inch for winograd neon3 8 years ago
  nihui c12fab569f fix convdw3x3s1 on aarch64 8 years ago
  nihui f133729c78 code style changes 8 years ago
  nihuini 03621aa7f9 more x86 stub for convolution and convolutiondepthwise 8 years ago
  Lamply 6612178960 correct arm convolution depthwise mistakes (#246) 8 years ago
  nihui 848c9a1ea7 code clean 8 years ago
  nihui 80fb28de90 unroll outch for convolution 3x3s1, about 10%~20% speed gain 8 years ago
  nihui df218110be unroll num_output for innerproduct, about 60% speed gain 8 years ago
  nihui aaa1ffcef0 emmmm, prefer w h 8 years ago
  nihui d68eb4cd15 wrap benchmark gettimeofday 8 years ago
  Linghan Cheung 811b6ba1b6 print benchmark information for every layer, especially for CONVOLUTION (#241) 8 years ago
  nihuini d2ee4e7d27 ld1 and st1 handle data endian mode per element 8 years ago
  nihui 08e261f423 innerproduct produce continous blob, fix #236 8 years ago
  nihui 682b0d3c0d prelu on vector and image 8 years ago
  nihui 14a2e23407 enable embed layer 8 years ago
  nihui c9789fb879 slice dim 8 years ago
  nihuini 67b80183dd fix param load using external memory 8 years ago
  nihuini 7fc23025d4 unroll outch for convolution 1x1 stride 2, about 15%~55% speed gain 8 years ago
  nihuini ccbb94d835 fix build 8 years ago
  nihuini e471028f53 fix avg pooling in tail pad 8 years ago
  nihuini a84ba8fc0f element type storage support in Mat, move data member the first so that a pointer to Mat is a pointer to data, convenient index access for float vector 8 years ago
  nihuini 8773035891 another implementation for winograd 3x3, about 15%~30% speed gains on small images 8 years ago
  nihui 2a62a98e1a allow constructing paramdict and modelbin from userside 8 years ago
  nihui 10b86c2af5 create layer from type name 8 years ago
  nihui 118e037f33 arm neon optimize for mat fill 8 years ago
  nihui 7a43c45e80 remove deprecated code 8 years ago
  nihui a181d25098 new model load api, fix #215 8 years ago
  nihuini b84ba31c23 enable light mode by default 8 years ago
  peng 5ac2de8963 fix shufflechannel 8 years ago
  nihuini df5e04260a fix conv1x1s1 bug 8 years ago
  nihuini 9280a068fe unroll outch for convolution 3x3 winograd64, reduce memory usage 8 years ago
  nihui 1f5c646ee0 pipeline optimize 8 years ago
  nihuini 0564021afc fix armv7 assembly 8 years ago
  nihuini 55ec189998 unroll outch for convolution 1x1 stride 1 8 years ago
  nihuini 57df1076ff neon optimize for depthwise convolution 3x3, about 20%~35% speed gain 8 years ago
  wind19870521 822214269d fix pooling2x2s2_max_neon stride bug 8 years ago
  zengping a54f14feca [fix-compile-warnings] fix compiler warnings, and add werror in CMakeLists.txt (#217) 8 years ago
  nihui 0f52418023 change input param order to w h c, replace caffe MemoryData to Input 8 years ago
  nihui 62913964d6 non-square kernel stride and padding w h for pooling 8 years ago
  nihui bdb70a2010 padding w h in convolution and deconvolution 8 years ago
  nihui 44b4519307 non-square convolution and deconvolution kernel stride dilation 8 years ago
  HustCoderHu 23a3254a7b fix memcpy error 8 years ago
  Zexin, Hu 81fb3818a5 add ShuffleChannel layer, only cpp, no arm yet (#210) 8 years ago
  nihuini 964040fe3c more runtime decisions for winograd path 8 years ago