关于降低模型体积
"提供量化脚本，对32位float转8位uint直接支持，模型体积量化后4M上下"
没有看到这个脚本？谢谢

如题，报错信息如下，
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ranlib: file: build/libmdl-static.a(mdl_jni.cpp.o) has no symbols
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ranlib build/libmdl-static.a
[ 89%] Built target mdl
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin/ranlib: file: build/libmdl-static.a(mdl_jni.cpp.o) has no symbols
[ 89%] Built target mdl-static
make: *** [all] Error 2

Thank you for this great work. Some feedback is as follows.

1. I have found that when my net has multi outputs, the program crashes. And I try to fix it in "mobile-deep-learning/src/net.cpp:87", "for(int i = start; i <= end; i++)" should be "for(int i = start; i < end; i++)".

2. The caffe2mdl tool does not support that input's width not equal to height. It is a limit that many tasks need to have different input width and height.

3. The predict output does not support get result by layer name and get result's shape. Should these features will be support in the future?

Best Regards,
WolffyChen

后面会增加层的支持吗？ 比如deconvolution layer prelu之类的

提供量化脚本，对32位float转8位uint直接支持，模型体积量化后4M上下
请问这个脚本在哪，怎么调用

Which ios device did you use to test the squeeze net? I installed the demo via QR on iphone7, and the model takes about 50ms per frame.

提供量化脚本，对32位float转8位uint直接支持，模型体积量化后4M上下
请问，上面这个脚本在哪里，怎么使用

Some mobile hardware platforms have only ubuntu system (or other linux systems), so is it possible to provide compiling config for ubuntu on arm (both armv7 and armv8), not only android and ios?

只看到在IOS GPU上，squeezenet能跑到30ms。能否提供在android上的性能，这样可以对比跟其他框架的性能。
从代码上来看，ncnn使用neon指令实现了convolution，感觉要比这里直接使用gemm要快一些。

when I run
./caffe2mdl model.prototxt model.caffemodel
it says Unknown bottom blob
tried several different models and get the same error

👍

请问自定义的模型转换后怎么放到so里面给Android使用？我从源码编译了Android的so，但是模型并没有用上。

百度的好工具，但是我想导入自己的图片库，比如识别各种手机，怎么导入呢？谢谢。

环境ubuntu16.04
/src/loader/loader.cpp:25:19: fatal error: zconf.h: No such file or directory
/src/commons/commons.cpp:96:48: error: ‘memcpy’ was not declared in this scope
/src/layer/pooling_layer.cpp:97:40: error: ‘INT_MAX’ was not declared in this scope

逻辑容易拆分的layer如pooling_layer,relu_layer,roi_pooling_layer,im2col，多线程比较简单；
看到gemm也没有使用多线程，其实可以考虑把openblas的sgemm nn和nt移植过来的，也不算复杂；
特定micro_kernel可以考虑纯汇编的unroll

squeezenet mdl without quantification loading fails ,
dump_without_quantification did not replace the character '/' in layer names.

执行./build.sh mac后：
./build.sh: line 24: cmake: command not found
make: *** No targets specified and no makefile found. Stop.

CMake Error at /usr/local/Cellar/cmake/3.5.1/share/cmake/Modules/FindPackageHandleStandardArgs.cmake:148 (message):
Could NOT find Protobuf (missing: PROTOBUF_LIBRARY PROTOBUF_INCLUDE_DIR)
Call Stack (most recent call first):
/usr/local/Cellar/cmake/3.5.1/share/cmake/Modules/FindPackageHandleStandardArgs.cmake:388 (_FPHSA_FAILURE_MESSAGE)
/usr/local/Cellar/cmake/3.5.1/share/cmake/Modules/FindProtobuf.cmake:308 (FIND_PACKAGE_HANDLE_STANDARD_ARGS)
tools/CMakeLists.txt:5 (find_package)

-- Configuring incomplete, errors occurred!
See also "/Users/icespring/mdl/mobile-deep-learning/build/release/x86/CMakeFiles/CMakeOutput.log".

which can solve by install protoc

求 Android GPU 支持。

BTW，已经看到 TODO 里面写了这个，使用什么方案支持 Android GPU 呢？

MTCNN等全卷积网络需要对模型的输入层做一下reshape。
我们这个框架怎么做到？

build.sh now only supports MAC compilation, need support Linux environment.

Hi guys,

It's cool to see that Baidu is making an open source library for doing deep learning on iOS and Android.

However, it seems that a fair amount of code in this project was taken from my Forge library. That is no problem in principle, as Forge is also open source and I'm happy that the source code I wrote helps other projects.

That said, I do hope you will respect the Forge license (https://github.com/hollance/Forge/blob/master/LICENSE.txt), which requires that you preserve the original license and copyright notice in the source code you have copied from Forge.

Thank you!

any plan for tensorflow model import?

When I use the caffe2mdl tools to convert ResNet model failed.
Did it support ResNet model ?

用ubuntu16.04编译，出现 this file was generated by a newer version of protoc which is，ubuntu安装的protoc是2.6.1，太旧了么？

Demo项目中加载我司的模型，跑到后面crash
m:1 n:6949 k:3200

#00 pc 00000000000310c4 /data/app/com.baidu.mdl.demo-1/lib/arm64/libmdl.so (mdl::Gemmer::pack_kxNR(int, float const*, int, int, float*)+24)
#1 pc 0000000000031194 /data/app/com.baidu.mdl.demo-1/lib/arm64/libmdl.so (mdl::Gemmer::pack_B(int, int, float const*, int, int, float*)+156)
#2 pc 0000000000031fa4 /data/app/com.baidu.mdl.demo-1/lib/arm64/libmdl.so (mdl::Gemmer::dgemm_nn(int, int, int, float, float const*, int, int, float const*, int, int, float, float*, int, int)+600)
#3 pc 000000000003216c /data/app/com.baidu.mdl.demo-1/lib/arm64/libmdl.so (mdl::Gemmer::sgemm(int, int, int, float const*, float const*, float*)+52)
#4 pc 00000000000258ac /data/app/com.baidu.mdl.demo-1/lib/arm64/libmdl.so (mdl::FCLayer::forward(int)+532)
#5 pc 0000000000034c18 /data/app/com.baidu.mdl.demo-1/lib/arm64/libmdl.so (mdl::Net::forward_from_to(float*, int, int, bool)+1172)
#6 pc 000000000003546c /data/app/com.baidu.mdl.demo-1/lib/arm64/libmdl.so (mdl::Net::predict(float*)+36)

使用3.4版本protobuf执行build.sh mac 报错
build/release/x86/tools/caffe.pb.h:17:2: error: #error This file was generated by an older version of protoc which is incompatible with your Protocol Buffer headers. Please regenerate this file with a newer version of protoc.

如题，
执行完脚本后，终端输出如下：
start running cycle : 0
load time : 70.449ms
total cost: 672.361ms.
89.6546 105.131 210.12 197.119
Done!
end running cycle : 0
，请问加粗的数字代表什么意思？是否可以图形可视化输出？

如题，
copy your model.prototxt and model.caffemodel to this path
also need the input data

./caffe2mdl model.prototxt model.caffemodel data

data是标注还是图像？

谢谢。

如题，
报错信息，如下，
[libprotobuf ERROR google/protobuf/text_format.cc:288] Error parsing text-format caffe.NetParameter: 743:24: Message type "caffe.LayerParameter" has no field named "smooth_l1_loss_param".
read_proto_from_text failed

谢谢！

我下载了example里面的Android代码,导入android studio中可以运行.但是拍照完了后没有方框??这是什么原因啊?

hi
thank you for the job！
I have a question below:
Does MDL optimized for mobile hardware Tx1 or only for phones?

thanks a lot!

My platform is running linux on firefly-3399 and I don't see an option in the build.sh

Thanks

TODO

Android GPU implementation
Converting Tensoflow Model to MDL

run the script './build.sh android'
get the error: CMake Error: Could not create named generator Android Gradle - Unix Makefiles
that's why?

Gemmer::sgemm not match caffe_cpu_gemm completely ,
when you test a model,you can found that the loss is iscrease.
then I found that Gemmer::sgemm not support the transformed mat.

can baidu solve it ?

Loading the Caffe model...
Traceback (most recent call last):
File "convert.py", line 12, in
import caffe_pb2
ImportError: bad magic number in 'caffe_pb2': b'\x03\xf3\r\n'

是否需要安装caffe？

I use ./build.sh android to build android ,
but there is a error
CMake Error: Could not create named generator Android Gradle - Unix Makefiles

who can solve this problem ?

I use ./build.sh android to build android ,
but there is a error
error This file requires compiler and library support for the ISO C++ 2011 standard. This support must be enabled with the -std=c++11 or -std=gnu++11 compiler options.
#error This file requires compiler and library support

who can solve it ?

It crashes when launch, iPhone 5s, iOS 11.

请问介绍中最后一条“将大量繁重的计算任务前置到overhead过程“ 中的 “overhead 过程“是指什么过程呢？谢谢

我想在cpu识别一张图片时使用squeezenet,使用test/model/squeezenet中提供的s_data.min.bin和s_model.min.json,demo崩溃,什么原因呢

Firstly, thank you for your hard work and contributions to the open source, but I found that the full name of CNN is wrong accidentally, would you mind correcting it? Thank you. Please refer to the attachment.