weiliangweiliang
ac7121f54b
fix the wrong variable name that cause wrong results. ( #741 )
* fix the wrong variable name that cause wrong results.
7 years ago
nihui
b75516b9b1
add the armv7a conv3x3s2, convdw3x3s1/s2 int8 implement without overflow : ) ( #738 )
* add the armv7a conv1x1s1 sgemm int8 implement without overflow : )
* add the armv7a conv3x3s2, convdw3x3s1/s2 int8 implement without overflow
7 years ago
weiliangweiliang
c79a6f7413
fix missing branch in arm 3x3s2 conv ( #737 )
7 years ago
BUG1989
229f8fd8db
add the armv7a conv3x3s2, convdw3x3s1/s2 int8 implement without overflow
7 years ago
nihuini
099189384f
fix load_param_bin, fix #732
7 years ago
nihui
f42901b6a5
update ios.toolchain.cmake and deploy on modern travis ci ( #719 )
* Update .travis.yml
* silence simulator warning
* fix ios arm detection
7 years ago
shujunhua
225d4925c0
fix lstm bug
7 years ago
BUG1989
4850eeed6f
add the armv7a conv1x1s1 sgemm int8 implement without overflow : ) ( #713 )
7 years ago
nihuini
0239d2ef48
another internal_nconsumed_ptr reset case, fix #706
7 years ago
nihuini
b2ffc339c0
reset internal_nconsumed_ptr before mem_scanf on msvc, fix #706
7 years ago
nihuini
fa3b8cfb87
print depthwise convolution and deconvolution kernel info in per-layer benchmark
7 years ago
nihuini
569be5a196
wrapper for yuv420sp bilinear resize
7 years ago
nihuini
1615f7de6c
the widely used yuv420sp2rgb, the fast approximate version
7 years ago
nihuini
49f5ac8f3d
implement resize bilinear c2
7 years ago
nihuini
60c40a8608
move pixel resize family to separated source file
7 years ago
nihuini
6a9ac1581e
fix clobber list in convdw5x5s1_neon and convdw5x5s2_neon
7 years ago
nihuini
69d2c48bb8
improve group deconvolution openmp scheduler
7 years ago
Eric Liu
9558c27daa
Fixed a yolov3 resolution bug ( #696 )
7 years ago
nihui
adfbb9eb25
fix caffe ssd, somewhat ugly though ...
7 years ago
Abdel Younes
b48ec83e36
fix: avoid rebuilding after cmake
7 years ago
nihuini
65be036aa6
mxnet-ssd done
7 years ago
nihuini
c77383f623
mxnet-ssd wip ...
7 years ago
Abdel Younes
a941701f98
fix: c++ warnings ( #666 )
7 years ago
nihuini
481722648c
fix ooops .... :]
7 years ago
nihuini
94e3a79ee9
fix crash in priorbox when min_sizes or max_sizes is empty
7 years ago
nihuini
0b34dd59fb
trival fix for the last element
7 years ago
nihuini
c44e49d3e9
implement roialign layer
7 years ago
Abdel Younes
a2611f4720
build: various cmake cleanups ( #657 )
* Those changes allows to use src/CMakeLists.txt directly
within a project that has its own toolchain and options.
* factorize arm and x86 cmake code,
* allow unix and android build for any kind of arm combination
* Introduce NCNN_CMAKE_VERBOSE option to reduce cmake verbosity
* use modern cmake for OpenMP detection and usage
7 years ago
nihuini
250f6b8bdd
fix build
7 years ago
nihuini
4837af4c25
initial effort for mxnet-ssd
7 years ago
nihuini
3a69f1c68b
implement psroipooling layer
7 years ago
nihuini
a5e57fa22e
implement multiscale yolov2, update example model comment
7 years ago
Jianjun Liu
8c79c90102
variable name should be remain.
7 years ago
karl2015
201ebb8e44
fix dequantize register error ( #653 )
* fix dequantize register error
* fix dequantize register error
7 years ago
nihuini
8fda293f91
neon optimize for depthwise convolution 5x5 :P
7 years ago
nihui
343524d29e
do not break layer type enum compability
7 years ago
Ming-Yang Liu
dfbed5a0ad
Support yolov3 and mobilenet yolov3
7 years ago
Corentin Barreau
736ed5e835
Support for Raspberry Pi 3 compilation ( #640 )
* Add: pi3.toolchain.cmake - CMake variables for Pi3 compilation.
* Modify: CMakeLists.txt - Add definitions for Pi3 compilation.
* Modify: src/cpu.cpp - Use getpid() if -PI3 is used.
7 years ago
BUG1989
971fe2f639
fix the bug of aarch64 conv1x1 int8 'asm' op less than 30 ( #628 )
* add aarch64 int8 neon implement
* fix the bug of aarch64 conv1x1 int8 'asm' op less than 30
* Update convolution_1x1_int8.h
7 years ago
RogerOu
0139939456
fix bug : undefined reference to 'typeinfo for ncnn::Allocator'
7 years ago
nihui
6afcd2299f
enable ncnn coverage
7 years ago
nihuini
4af3ec223b
fix clip arm
7 years ago
nihuini
4e68a29eff
fix build on msvc, second try
7 years ago
nihuini
4a9e452cb5
fix build on msvc, first try
7 years ago
nihuini
9fbaf7fb0b
coding style changes
7 years ago
BUG1989
c171db255f
update aarch64 conv1x1s1 int8, speedup 10%.
7 years ago
BUG1989
08b39dffb0
add aarch64 int8 neon implement
7 years ago
RogerOu
add45371de
fix warning: allocator non-virtual destructor
7 years ago
RogerOu
0633973bfa
clip layer support arm neon
7 years ago
nihuini
8eeea9ef01
fix reduction dim 2, fix #593
7 years ago