Hi Haoshu Fang,
Thank you for sharing the code. But when I ran the "python examples/rmpe/demo.py" without runing the "jupyter notebook examples/rmpe/Regional\ Multi-person\ Pose\ Estimation.ipynb",
the output was "NameError: name 'PCK_pose_NMS' is not defined". Then I checked your code in whole project about this function named "PCK_pose_NMS", and I did not find it.
Did I miss something? Any suggestions would be helpful to me.
Thanks!!!

Samylee

hi ，I have a very strange problem, that is, I did not get an error during the running of the program, and then automatically stopped, can you help me? thanks.
.....
.....
I1121 23:31:05.171264 6824 net.cpp:228] pool1s does not need backward computation.
I1121 23:31:05.171267 6824 net.cpp:228] conv1_relu does not need backward computation.
I1121 23:31:05.171270 6824 net.cpp:228] conv1_scale does not need backward computation.
I1121 23:31:05.171273 6824 net.cpp:228] conv1_bn does not need backward computation.
I1121 23:31:05.171277 6824 net.cpp:228] conv1 does not need backward computation.
I1121 23:31:05.171280 6824 net.cpp:228] data_input_0_split does not need backward computation.
I1121 23:31:05.171284 6824 net.cpp:228] input does not need backward computation.
I1121 23:31:05.171286 6824 net.cpp:270] This network produces output prediction_heatmap
I1121 23:31:05.171772 6824 net.cpp:283] Network initialization done.
/home/h/anaconda3/envs/py2/lib/python2.7/site-packages/skimage/io/_io.py:49: UserWarning: as_grey has been deprecated in favor of as_gray
warn('as_grey has been deprecated in favor of as_gray')
/home/h/anaconda3/envs/py2/lib/python2.7/site-packages/skimage/transform/_warps.py:105: UserWarning: The default mode, 'constant', will be changed to 'reflect' in skimage 0.15.
warn("The default mode, 'constant', will be changed to 'reflect' in "
/home/h/anaconda3/envs/py2/lib/python2.7/site-packages/skimage/transform/_warps.py:110: UserWarning: Anti-aliasing will be enabled by default in skimage 0.15 to avoid aliasing artifacts when down-sampling images.
warn("Anti-aliasing will be enabled by default in skimage 0.15 to "
(py2) h@h-pc:~/RMPE/RMPE$

Dear Hao-Shu Fang:
My name is TianYu Zhu，My email is [email protected]. I have sent you an email for the code. Did you receive it？Looking forward to your reply!

Hi, Haoshu
Sorry to bother you. I tried a lot but failed to deal with this problem.
Everytime I use the command--make, it showed this error. So I can't compile the code successfully. ( It is ok to compile the caffe code contributed by BVLC. )
/usr/include/boost/property_tree/detail/json_parser_read.hpp: In constructor‘boost::property_tree::json_parser::json_grammar::definition::definition(constboost::property_tree::json_parser::json_grammar&)':
/usr/include/boost/property_tree/detail/json_parser_read.hpp:257:264: error: 'type name'declared as function returning an array
/usr/include/boost/property_tree/detail/json_parser_read.hpp:257:264: error: 'type name'declared as function returning an array
make: *** [.build_release/cuda/src/caffe/layers/detection_output_layer.o] Error 1
make: *** Waiting for unfinished jobs....

I find your train.txt is different from caffe-heatmap-master where heatmap definition?how to make ?

您好，学长：
非常感谢您分享的程序。在Installation步骤中，我按照您所写执行，每当执行make runtest -j8时，总会报这样一个错误：
[----------] 2 tests from HardSpatialTransformerLayerTest/1, where TypeParam = caffe::CPUDevice

[ RUN ] HardSpatialTransformerLayerTest/1.TestGradient

	Spatial Transformer Layer:: LayerSetUp: 	Getting output_H_ and output_W_

	Spatial Transformer Layer:: LayerSetUp: 	output_H_ = 7, output_W_ = 7

	Spatial Transformer Layer:: LayerSetUp: 	Getting pre-defined parameters

	Spatial Transformer Layer:: LayerSetUp: 	Initializing the matrix for output grid

	Spatial Transformer Layer:: LayerSetUp: 	Initializing the matrix for input grid

	Spatial Transformer Layer:: LayerSetUp: 	Initialization finished.

F1217 14:07:56.563449 21562 st_layer.cpp:263] Check failed: false Don't use the CPU implementation! If you really want to, delete the CHECK in st_layer.cpp file. Line number: 242-243. De_transform is not supported.

*** Check failure stack trace: ***

@     0x7f7d5fb6de6d  (unknown)

@     0x7f7d5fb6fced  (unknown)

@     0x7f7d5fb6da5c  (unknown)

@     0x7f7d5fb7063e  (unknown)

@     0x7f7d5b6e1119  caffe::SpatialTransformerLayer<>::Forward_cpu()

@           0x48402a  caffe::Layer<>::Forward()

@           0x4cfe3b  caffe::GradientChecker<>::CheckGradientSingle()

@           0x4d0a7b  caffe::GradientChecker<>::CheckGradientExhaustive()

@           0x95f672  caffe::HardSpatialTransformerLayerTest_TestGradient_Test<>::TestBody()

@           0x9b2ce3  testing::internal::HandleExceptionsInMethodIfSupported<>()

@           0x9aaae7  testing::Test::Run()

@           0x9aab8e  testing::TestInfo::Run()

@           0x9aac95  testing::TestCase::Run()

@           0x9acf28  testing::internal::UnitTestImpl::RunAllTests()

@           0x9ad1b7  testing::UnitTest::Run()

@           0x475aaf  main

@     0x7f7d5a884b15  __libc_start_main

@           0x47e909  (unknown)

make: *** [runtest] Aborted (core dumped)

我在Makefile.config中仅修改了BLAS := mkl 和 WITH_PYTHON_LAYER := 1。
另外在make runtest开始时会出现找不到fish-bike.jpg的错误信息，我只能从我们服务器的caffe/examples/images中复制这张图片到RMPE/examples/images中。
是不是我某些地方做的不太对？希望您能给予帮助。非常感谢！！

siyilingting

Which opencv version do you support?
My OS is ubuntu 16.04 and I built opencv 3.1.0 in my computer, when I build the code an error as follow:
build_release/lib/libcaffe.so：对‘cv::VideoCapture::set(int, double)’未定义的引用
and there are some same contents omitted.

Is PGPG placed after human detector and before SSTN?

Hi Haoshu Fang,
Thank you for sharing the code.And Sorry to bother you.
In this project. If I want to see the performance about the person detection, what can i do for the code.Is there any direct test code?

Hi Haoshu,

In the process of evaluating the model, it seems that the function 'set_pred()' does not exist. However, there is one function named 'setpred()'. Are those two functions the same?

Thanks

I think the description of pose_NMS in the paper is wrong, d(Pi, Pj|Λ, λ) ≤ η should be d(Pi, Pj|Λ, λ) > η???
The code seems to support my idea, too! see below,
delete_ids = np.arange(img_scores.shape[0])[simi > gamma]

Please give me the correct answer, thanks very much!

Hi, Haoshu thanks for you code !

Sorry to bother you.

when i run python examples/rmpe/demo.py
I meet this problem. could you give some advice？

I0713 08:45:44.166777 27209 net.cpp:270] This network produces output prediction_heatmap
I0713 08:45:44.167436 27209 net.cpp:283] Network initialization done.
*** Error in `python': munmap_chunk(): invalid pointer: 0x00000000023af490 ***

======= Backtrace: =========
/lib/x86_64-linux-gnu/libc.so.6(+0x777e5)[0x7f00c734e7e5]
/lib/x86_64-linux-gnu/libc.so.6(cfree+0x1a8)[0x7f00c735b698]
/usr/local/lib/libprotobuf.so.9(_ZN6google8protobuf8internal28DestroyDefaultRepeatedFieldsEv+0x1f)[0x7f00a103ab5f]
/usr/local/lib/libprotobuf.so.9(_ZN6google8protobuf23ShutdownProtobufLibraryEv+0x8b)[0x7f00a1039f4b]
/usr/lib/x86_64-linux-gnu/libmirprotobuf.so.3(+0x20329)[0x7f007da6d329]
/lib64/ld-linux-x86-64.so.2(+0x10de7)[0x7f00c78cede7]
/lib/x86_64-linux-gnu/libc.so.6(+0x39ff8)[0x7f00c7310ff8]
/lib/x86_64-linux-gnu/libc.so.6(+0x3a045)[0x7f00c7311045]
/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf7)[0x7f00c72f7837]
python(_start+0x29)[0x49d9d9]
======= Memory map: ========

师兄您好，
mpii数据集上，有些点的坐标是在图片范围之外的，还有些点的坐标直接是（0，0），请问这些点的groundtruth在生成heatmap的时候，是如何处理的？

I tested the demo on many images. The processing speed is relatively fast, but usually some parts of the predicted skeleton is somewhat deviated from the body to the upper left. Why is that?

When I run make runtest -j 8, the error comes out like this. But I cannot find libcudart.so.5.5 in cuda's lib (/usr/local/cuda/lib64). My cuda version is 8.0. Is this problem caused by version incompatibility?