Py-faster-rcnn: Tried to run demo.py inside tools folder but the detection accuracy has been very low

Created on 28 Oct 2016 · 6Comments · Source: rbgirshick/py-faster-rcnn

I just download this repo as the tutorial said and tried to run the demo.py script but the results are quite weird, the accuracy is low and the score typically ranges from 0 to 0.05. Is the model not trained well?

Source

xinleipan

Most helpful comment

@neoscode Sure. I've finally solved this problem and want to share with the public in case anyone who has the same problems. So the operating system is Ubuntu 14.04,with Titan X Pascal (the newest GPU).

First step: Install CUDA 8.0 and CUDNN 5.0 for CUDA 8.0: https://www.pugetsystems.com/labs/hpc/Install-Ubuntu-16-04-or-14-04-and-CUDA-8-and-7-5-for-NVIDIA-Pascal-GPU-825/
Download Faster-RCNN from this repo. Before you make caffe, do these steps first:

(1) cd into caffe-fast-rcnn, and type in "git remote add caffe https://github.com/BVLC/caffe.git"
(2) "git fetch caffe"
(3) "git merge caffe/master"

Still before you make caffe, comment the line "self_.attr("phase") = static_cast(this->phase_);" in /caffe-fast-rcnn/include/caffe/layers/python_layer.hpp.
Then revise your Makefile.config accordingly. I mean uncomment "USE_CUDNN :=1" and WITH_PYTHON_LAYER :=1.

Revise INCLUDE_DIRS as:
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial/

revise LIBRARY_DIRS as:
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu/hdf5/serial/

revise PYTHON_INCLUDE as:
PYTHON_INCLUDE := /usr/include/python2.7 \
/usr/lib/python2.7/dist-packages/numpy/core/include \
/usr/local/lib/python2.7/dist-packages/numpy/core/include
Then "make -j8" and "make pycaffe".

YOU ARE ALL SET!!!
Then you can check your demo.py. ( Remember to download your models first by doing "./data/scripts/fetch_faster_rcnn_models.sh".

Hope it helps!

xinleipan on 20 Nov 2016

👍2

All 6 comments

checked by 16-11-2016 using cuda 8.0

step 1: downgrade cudnnv5 to cudnnv4
download cudnnv4 then

sudo cp cuda/lib64/lib* /usr/local/cuda-8.0/lib64/
sudo cp cuda/include/cudnn.h /usr/local/cuda-8.0/include/
echo 'export LD_LIBRARY_PATH=/usr/local/cuda-8.0/lib64:$LD_LIBRARY_PATH' >> ~/.bashrc

step 2:enable cudnn in Makefile.config
inside Makefile.config comment the follow lines
USE_CUDNN := 1
WITH_PYTHON_LAYER := 1
and replace the respective lines by
CUDA_DIR := /usr/local/cuda-8.0
INCLUDE_DIRS := $(PYTHON_INCLUDE) /usr/local/include /usr/include/hdf5/serial
LIBRARY_DIRS := $(PYTHON_LIB) /usr/local/lib /usr/lib /usr/lib/x86_64-linux-gnu /usr/lib/x86_64-linux-gnu/hdf5/serial

make -j8 && make pycaffe
done!

If this solve your problem pls like this comment and follow my hub

max-ng on 15 Nov 2016

+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| 0 1215 G /usr/lib/xorg/Xorg 155MiB |
| 0 1757 G compiz 129MiB |
+-----------------------------------------------------------------------------+

I used Ubuntu 14.04, with CUDA 7.5 for ubuntu 14.04, cudnn 4.0 for linux. Installed according to https://www.pugetsystems.com/labs/hpc/Install-Ubuntu-16-04-or-14-04-and-CUDA-8-and-7-5-for-NVIDIA-Pascal-GPU-825/
Then I followed the step in this git repo to install caffe. However, the output when running " python demo.py --gpu 0 --net zf" is:
...
}
layer {
name: "roi_pool5"
type: "ROIPooling"
bottom: "conv5_3"
bottom: "rois"
top: "pool5"
roi_pooling_param {
pooled_h: 7
pooled_w: 7
spatial_scale: 0.0625
}
}
layer {
name: "fc6"
type: "InnerProduct"
bottom: "pool5"
top: "fc6"
inner_product_param {
num_output: 4096
}
}
layer {
name: "relu6"
type: "ReLU"
bottom: "fc6"
top: "fc6"
}
layer {
name: "fc7"
type: "InnerProduct"
bottom: "fc6"
top: "fc7"
inner_product_param {
num_output: 4096
}
}
layer {
name: "relu7"
type: "ReLU"
bottom: "fc7"
top: "fc7"
}
layer {
name: "cls_score"
type: "InnerProduct"
bottom: "fc7"
top: "cls_score"
inner_product_param {
num_output: 21
}
}
layer {
name: "bbox_pred"
type: "InnerProduct"
bottom: "fc7"
top: "bbox_pred"
inner_product_param {
num_output: 84
}
}
layer {
name: "cls_prob"
type: "Softmax"
bottom: "cls_score"
top: "cls_prob"
}
I1117 20:11:00.964956 1012 net.cpp:413] Input 0 -> data
I1117 20:11:00.974936 1012 net.cpp:413] Input 1 -> im_info
I1117 20:11:00.974999 1012 layer_factory.hpp:77] Creating layer conv1_1
I1117 20:11:00.975028 1012 net.cpp:106] Creating Layer conv1_1
I1117 20:11:00.975030 1012 net.cpp:454] conv1_1 <- data
I1117 20:11:00.975049 1012 net.cpp:411] conv1_1 -> conv1_1
I1117 20:11:01.206707 1012 net.cpp:150] Setting up conv1_1
I1117 20:11:01.206723 1012 net.cpp:157] Top shape: 1 64 224 224 (3211264)
I1117 20:11:01.206727 1012 net.cpp:165] Memory required for data: 12845056
I1117 20:11:01.206751 1012 layer_factory.hpp:77] Creating layer relu1_1
I1117 20:11:01.206759 1012 net.cpp:106] Creating Layer relu1_1
I1117 20:11:01.206763 1012 net.cpp:454] relu1_1 <- conv1_1
I1117 20:11:01.206766 1012 net.cpp:397] relu1_1 -> conv1_1 (in-place)
I1117 20:11:01.206933 1012 net.cpp:150] Setting up relu1_1
I1117 20:11:01.206938 1012 net.cpp:157] Top shape: 1 64 224 224 (3211264)
I1117 20:11:01.206940 1012 net.cpp:165] Memory required for data: 25690112
I1117 20:11:01.206943 1012 layer_factory.hpp:77] Creating layer conv1_2
I1117 20:11:01.206948 1012 net.cpp:106] Creating Layer conv1_2
I1117 20:11:01.206950 1012 net.cpp:454] conv1_2 <- conv1_1
I1117 20:11:01.206954 1012 net.cpp:411] conv1_2 -> conv1_2
I1117 20:11:01.207641 1012 net.cpp:150] Setting up conv1_2
I1117 20:11:01.207649 1012 net.cpp:157] Top shape: 1 64 224 224 (3211264)
I1117 20:11:01.207651 1012 net.cpp:165] Memory required for data: 38535168
I1117 20:11:01.207657 1012 layer_factory.hpp:77] Creating layer relu1_2
I1117 20:11:01.207675 1012 net.cpp:106] Creating Layer relu1_2
I1117 20:11:01.207679 1012 net.cpp:454] relu1_2 <- conv1_2
I1117 20:11:01.207681 1012 net.cpp:397] relu1_2 -> conv1_2 (in-place)
I1117 20:11:01.207834 1012 net.cpp:150] Setting up relu1_2
I1117 20:11:01.207839 1012 net.cpp:157] Top shape: 1 64 224 224 (3211264)
I1117 20:11:01.207840 1012 net.cpp:165] Memory required for data: 51380224
I1117 20:11:01.207842 1012 layer_factory.hpp:77] Creating layer pool1
I1117 20:11:01.207846 1012 net.cpp:106] Creating Layer pool1
I1117 20:11:01.207849 1012 net.cpp:454] pool1 <- conv1_2
I1117 20:11:01.207851 1012 net.cpp:411] pool1 -> pool1
I1117 20:11:01.207906 1012 net.cpp:150] Setting up pool1
I1117 20:11:01.207922 1012 net.cpp:157] Top shape: 1 64 112 112 (802816)
I1117 20:11:01.207924 1012 net.cpp:165] Memory required for data: 54591488
I1117 20:11:01.207926 1012 layer_factory.hpp:77] Creating layer conv2_1
I1117 20:11:01.207931 1012 net.cpp:106] Creating Layer conv2_1
I1117 20:11:01.207932 1012 net.cpp:454] conv2_1 <- pool1
I1117 20:11:01.207936 1012 net.cpp:411] conv2_1 -> conv2_1
I1117 20:11:01.209638 1012 net.cpp:150] Setting up conv2_1
I1117 20:11:01.209646 1012 net.cpp:157] Top shape: 1 128 112 112 (1605632)
I1117 20:11:01.209648 1012 net.cpp:165] Memory required for data: 61014016
I1117 20:11:01.209655 1012 layer_factory.hpp:77] Creating layer relu2_1
I1117 20:11:01.209673 1012 net.cpp:106] Creating Layer relu2_1
I1117 20:11:01.209676 1012 net.cpp:454] relu2_1 <- conv2_1
I1117 20:11:01.209679 1012 net.cpp:397] relu2_1 -> conv2_1 (in-place)
I1117 20:11:01.209841 1012 net.cpp:150] Setting up relu2_1
I1117 20:11:01.209846 1012 net.cpp:157] Top shape: 1 128 112 112 (1605632)
I1117 20:11:01.209847 1012 net.cpp:165] Memory required for data: 67436544
I1117 20:11:01.209849 1012 layer_factory.hpp:77] Creating layer conv2_2
I1117 20:11:01.209853 1012 net.cpp:106] Creating Layer conv2_2
I1117 20:11:01.209856 1012 net.cpp:454] conv2_2 <- conv2_1
I1117 20:11:01.209873 1012 net.cpp:411] conv2_2 -> conv2_2
I1117 20:11:01.210531 1012 net.cpp:150] Setting up conv2_2
I1117 20:11:01.210539 1012 net.cpp:157] Top shape: 1 128 112 112 (1605632)
I1117 20:11:01.210541 1012 net.cpp:165] Memory required for data: 73859072
I1117 20:11:01.210546 1012 layer_factory.hpp:77] Creating layer relu2_2
I1117 20:11:01.210564 1012 net.cpp:106] Creating Layer relu2_2
I1117 20:11:01.210566 1012 net.cpp:454] relu2_2 <- conv2_2
I1117 20:11:01.210571 1012 net.cpp:397] relu2_2 -> conv2_2 (in-place)
I1117 20:11:01.210875 1012 net.cpp:150] Setting up relu2_2
I1117 20:11:01.210880 1012 net.cpp:157] Top shape: 1 128 112 112 (1605632)
I1117 20:11:01.210882 1012 net.cpp:165] Memory required for data: 80281600
I1117 20:11:01.210886 1012 layer_factory.hpp:77] Creating layer pool2
I1117 20:11:01.210889 1012 net.cpp:106] Creating Layer pool2
I1117 20:11:01.210891 1012 net.cpp:454] pool2 <- conv2_2
I1117 20:11:01.210909 1012 net.cpp:411] pool2 -> pool2
I1117 20:11:01.210961 1012 net.cpp:150] Setting up pool2
I1117 20:11:01.210965 1012 net.cpp:157] Top shape: 1 128 56 56 (401408)
I1117 20:11:01.210968 1012 net.cpp:165] Memory required for data: 81887232
I1117 20:11:01.210968 1012 layer_factory.hpp:77] Creating layer conv3_1
I1117 20:11:01.210973 1012 net.cpp:106] Creating Layer conv3_1
I1117 20:11:01.210974 1012 net.cpp:454] conv3_1 <- pool2
I1117 20:11:01.210978 1012 net.cpp:411] conv3_1 -> conv3_1
I1117 20:11:01.211725 1012 net.cpp:150] Setting up conv3_1
I1117 20:11:01.211732 1012 net.cpp:157] Top shape: 1 256 56 56 (802816)
I1117 20:11:01.211735 1012 net.cpp:165] Memory required for data: 85098496
I1117 20:11:01.211740 1012 layer_factory.hpp:77] Creating layer relu3_1
I1117 20:11:01.211758 1012 net.cpp:106] Creating Layer relu3_1
I1117 20:11:01.211760 1012 net.cpp:454] relu3_1 <- conv3_1
I1117 20:11:01.211765 1012 net.cpp:397] relu3_1 -> conv3_1 (in-place)
I1117 20:11:01.211885 1012 net.cpp:150] Setting up relu3_1
I1117 20:11:01.211890 1012 net.cpp:157] Top shape: 1 256 56 56 (802816)
I1117 20:11:01.211892 1012 net.cpp:165] Memory required for data: 88309760
I1117 20:11:01.211894 1012 layer_factory.hpp:77] Creating layer conv3_2
I1117 20:11:01.211897 1012 net.cpp:106] Creating Layer conv3_2
I1117 20:11:01.211899 1012 net.cpp:454] conv3_2 <- conv3_1
I1117 20:11:01.211902 1012 net.cpp:411] conv3_2 -> conv3_2
I1117 20:11:01.213220 1012 net.cpp:150] Setting up conv3_2
I1117 20:11:01.213228 1012 net.cpp:157] Top shape: 1 256 56 56 (802816)
I1117 20:11:01.213230 1012 net.cpp:165] Memory required for data: 91521024
I1117 20:11:01.213235 1012 layer_factory.hpp:77] Creating layer relu3_2
I1117 20:11:01.213238 1012 net.cpp:106] Creating Layer relu3_2
I1117 20:11:01.213255 1012 net.cpp:454] relu3_2 <- conv3_2
I1117 20:11:01.213259 1012 net.cpp:397] relu3_2 -> conv3_2 (in-place)
I1117 20:11:01.213394 1012 net.cpp:150] Setting up relu3_2
I1117 20:11:01.213399 1012 net.cpp:157] Top shape: 1 256 56 56 (802816)
I1117 20:11:01.213402 1012 net.cpp:165] Memory required for data: 94732288
I1117 20:11:01.213403 1012 layer_factory.hpp:77] Creating layer conv3_3
I1117 20:11:01.213407 1012 net.cpp:106] Creating Layer conv3_3
I1117 20:11:01.213409 1012 net.cpp:454] conv3_3 <- conv3_2
I1117 20:11:01.213428 1012 net.cpp:411] conv3_3 -> conv3_3
I1117 20:11:01.214766 1012 net.cpp:150] Setting up conv3_3
I1117 20:11:01.214774 1012 net.cpp:157] Top shape: 1 256 56 56 (802816)
I1117 20:11:01.214776 1012 net.cpp:165] Memory required for data: 97943552
I1117 20:11:01.214781 1012 layer_factory.hpp:77] Creating layer relu3_3
I1117 20:11:01.214784 1012 net.cpp:106] Creating Layer relu3_3
I1117 20:11:01.214787 1012 net.cpp:454] relu3_3 <- conv3_3
I1117 20:11:01.214789 1012 net.cpp:397] relu3_3 -> conv3_3 (in-place)
I1117 20:11:01.214907 1012 net.cpp:150] Setting up relu3_3
I1117 20:11:01.214912 1012 net.cpp:157] Top shape: 1 256 56 56 (802816)
I1117 20:11:01.214915 1012 net.cpp:165] Memory required for data: 101154816
I1117 20:11:01.214915 1012 layer_factory.hpp:77] Creating layer pool3
I1117 20:11:01.214920 1012 net.cpp:106] Creating Layer pool3
I1117 20:11:01.214922 1012 net.cpp:454] pool3 <- conv3_3
I1117 20:11:01.214941 1012 net.cpp:411] pool3 -> pool3
I1117 20:11:01.214977 1012 net.cpp:150] Setting up pool3
I1117 20:11:01.214980 1012 net.cpp:157] Top shape: 1 256 28 28 (200704)
I1117 20:11:01.214982 1012 net.cpp:165] Memory required for data: 101957632
I1117 20:11:01.214983 1012 layer_factory.hpp:77] Creating layer conv4_1
I1117 20:11:01.214987 1012 net.cpp:106] Creating Layer conv4_1
I1117 20:11:01.214989 1012 net.cpp:454] conv4_1 <- pool3
I1117 20:11:01.214993 1012 net.cpp:411] conv4_1 -> conv4_1
I1117 20:11:01.217676 1012 net.cpp:150] Setting up conv4_1
I1117 20:11:01.217690 1012 net.cpp:157] Top shape: 1 512 28 28 (401408)
I1117 20:11:01.217692 1012 net.cpp:165] Memory required for data: 103563264
I1117 20:11:01.217700 1012 layer_factory.hpp:77] Creating layer relu4_1
I1117 20:11:01.217720 1012 net.cpp:106] Creating Layer relu4_1
I1117 20:11:01.217725 1012 net.cpp:454] relu4_1 <- conv4_1
I1117 20:11:01.217730 1012 net.cpp:397] relu4_1 -> conv4_1 (in-place)
I1117 20:11:01.217855 1012 net.cpp:150] Setting up relu4_1
I1117 20:11:01.217874 1012 net.cpp:157] Top shape: 1 512 28 28 (401408)
I1117 20:11:01.217876 1012 net.cpp:165] Memory required for data: 105168896
I1117 20:11:01.217878 1012 layer_factory.hpp:77] Creating layer conv4_2
I1117 20:11:01.217897 1012 net.cpp:106] Creating Layer conv4_2
I1117 20:11:01.217900 1012 net.cpp:454] conv4_2 <- conv4_1
I1117 20:11:01.217903 1012 net.cpp:411] conv4_2 -> conv4_2
I1117 20:11:01.222105 1012 net.cpp:150] Setting up conv4_2
I1117 20:11:01.222120 1012 net.cpp:157] Top shape: 1 512 28 28 (401408)
I1117 20:11:01.222122 1012 net.cpp:165] Memory required for data: 106774528
I1117 20:11:01.222146 1012 layer_factory.hpp:77] Creating layer relu4_2
I1117 20:11:01.222165 1012 net.cpp:106] Creating Layer relu4_2
I1117 20:11:01.222168 1012 net.cpp:454] relu4_2 <- conv4_2
I1117 20:11:01.222172 1012 net.cpp:397] relu4_2 -> conv4_2 (in-place)
I1117 20:11:01.222349 1012 net.cpp:150] Setting up relu4_2
I1117 20:11:01.222353 1012 net.cpp:157] Top shape: 1 512 28 28 (401408)
I1117 20:11:01.222355 1012 net.cpp:165] Memory required for data: 108380160
I1117 20:11:01.222357 1012 layer_factory.hpp:77] Creating layer conv4_3
I1117 20:11:01.222362 1012 net.cpp:106] Creating Layer conv4_3
I1117 20:11:01.222364 1012 net.cpp:454] conv4_3 <- conv4_2
I1117 20:11:01.222380 1012 net.cpp:411] conv4_3 -> conv4_3
I1117 20:11:01.225921 1012 net.cpp:150] Setting up conv4_3
I1117 20:11:01.225939 1012 net.cpp:157] Top shape: 1 512 28 28 (401408)
I1117 20:11:01.225941 1012 net.cpp:165] Memory required for data: 109985792
I1117 20:11:01.225947 1012 layer_factory.hpp:77] Creating layer relu4_3
I1117 20:11:01.225953 1012 net.cpp:106] Creating Layer relu4_3
I1117 20:11:01.225956 1012 net.cpp:454] relu4_3 <- conv4_3
I1117 20:11:01.225960 1012 net.cpp:397] relu4_3 -> conv4_3 (in-place)
I1117 20:11:01.226248 1012 net.cpp:150] Setting up relu4_3
I1117 20:11:01.226255 1012 net.cpp:157] Top shape: 1 512 28 28 (401408)
I1117 20:11:01.226258 1012 net.cpp:165] Memory required for data: 111591424
I1117 20:11:01.226259 1012 layer_factory.hpp:77] Creating layer pool4
I1117 20:11:01.226263 1012 net.cpp:106] Creating Layer pool4
I1117 20:11:01.226265 1012 net.cpp:454] pool4 <- conv4_3
I1117 20:11:01.226269 1012 net.cpp:411] pool4 -> pool4
I1117 20:11:01.226310 1012 net.cpp:150] Setting up pool4
I1117 20:11:01.226326 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.226328 1012 net.cpp:165] Memory required for data: 111992832
I1117 20:11:01.226330 1012 layer_factory.hpp:77] Creating layer conv5_1
I1117 20:11:01.226335 1012 net.cpp:106] Creating Layer conv5_1
I1117 20:11:01.226336 1012 net.cpp:454] conv5_1 <- pool4
I1117 20:11:01.226339 1012 net.cpp:411] conv5_1 -> conv5_1
I1117 20:11:01.229813 1012 net.cpp:150] Setting up conv5_1
I1117 20:11:01.229830 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.229832 1012 net.cpp:165] Memory required for data: 112394240
I1117 20:11:01.229838 1012 layer_factory.hpp:77] Creating layer relu5_1
I1117 20:11:01.229845 1012 net.cpp:106] Creating Layer relu5_1
I1117 20:11:01.229847 1012 net.cpp:454] relu5_1 <- conv5_1
I1117 20:11:01.229851 1012 net.cpp:397] relu5_1 -> conv5_1 (in-place)
I1117 20:11:01.229980 1012 net.cpp:150] Setting up relu5_1
I1117 20:11:01.229985 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.229987 1012 net.cpp:165] Memory required for data: 112795648
I1117 20:11:01.229990 1012 layer_factory.hpp:77] Creating layer conv5_2
I1117 20:11:01.229995 1012 net.cpp:106] Creating Layer conv5_2
I1117 20:11:01.229996 1012 net.cpp:454] conv5_2 <- conv5_1
I1117 20:11:01.230000 1012 net.cpp:411] conv5_2 -> conv5_2
I1117 20:11:01.233445 1012 net.cpp:150] Setting up conv5_2
I1117 20:11:01.233461 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.233464 1012 net.cpp:165] Memory required for data: 113197056
I1117 20:11:01.233485 1012 layer_factory.hpp:77] Creating layer relu5_2
I1117 20:11:01.233491 1012 net.cpp:106] Creating Layer relu5_2
I1117 20:11:01.233495 1012 net.cpp:454] relu5_2 <- conv5_2
I1117 20:11:01.233500 1012 net.cpp:397] relu5_2 -> conv5_2 (in-place)
I1117 20:11:01.233616 1012 net.cpp:150] Setting up relu5_2
I1117 20:11:01.233621 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.233623 1012 net.cpp:165] Memory required for data: 113598464
I1117 20:11:01.233625 1012 layer_factory.hpp:77] Creating layer conv5_3
I1117 20:11:01.233630 1012 net.cpp:106] Creating Layer conv5_3
I1117 20:11:01.233633 1012 net.cpp:454] conv5_3 <- conv5_2
I1117 20:11:01.233650 1012 net.cpp:411] conv5_3 -> conv5_3
I1117 20:11:01.237792 1012 net.cpp:150] Setting up conv5_3
I1117 20:11:01.237808 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.237812 1012 net.cpp:165] Memory required for data: 113999872
I1117 20:11:01.237818 1012 layer_factory.hpp:77] Creating layer relu5_3
I1117 20:11:01.237823 1012 net.cpp:106] Creating Layer relu5_3
I1117 20:11:01.237825 1012 net.cpp:454] relu5_3 <- conv5_3
I1117 20:11:01.237829 1012 net.cpp:397] relu5_3 -> conv5_3 (in-place)
I1117 20:11:01.237959 1012 net.cpp:150] Setting up relu5_3
I1117 20:11:01.237964 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.237967 1012 net.cpp:165] Memory required for data: 114401280
I1117 20:11:01.237968 1012 layer_factory.hpp:77] Creating layer conv5_3_relu5_3_0_split
I1117 20:11:01.237977 1012 net.cpp:106] Creating Layer conv5_3_relu5_3_0_split
I1117 20:11:01.237978 1012 net.cpp:454] conv5_3_relu5_3_0_split <- conv5_3
I1117 20:11:01.237982 1012 net.cpp:411] conv5_3_relu5_3_0_split -> conv5_3_relu5_3_0_split_0
I1117 20:11:01.237985 1012 net.cpp:411] conv5_3_relu5_3_0_split -> conv5_3_relu5_3_0_split_1
I1117 20:11:01.238039 1012 net.cpp:150] Setting up conv5_3_relu5_3_0_split
I1117 20:11:01.238042 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.238044 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.238046 1012 net.cpp:165] Memory required for data: 115204096
I1117 20:11:01.238049 1012 layer_factory.hpp:77] Creating layer rpn_conv/3x3
I1117 20:11:01.238054 1012 net.cpp:106] Creating Layer rpn_conv/3x3
I1117 20:11:01.238055 1012 net.cpp:454] rpn_conv/3x3 <- conv5_3_relu5_3_0_split_0
I1117 20:11:01.238072 1012 net.cpp:411] rpn_conv/3x3 -> rpn/output
I1117 20:11:01.241701 1012 net.cpp:150] Setting up rpn_conv/3x3
I1117 20:11:01.241719 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.241722 1012 net.cpp:165] Memory required for data: 115605504
I1117 20:11:01.241729 1012 layer_factory.hpp:77] Creating layer rpn_relu/3x3
I1117 20:11:01.241737 1012 net.cpp:106] Creating Layer rpn_relu/3x3
I1117 20:11:01.241740 1012 net.cpp:454] rpn_relu/3x3 <- rpn/output
I1117 20:11:01.241745 1012 net.cpp:397] rpn_relu/3x3 -> rpn/output (in-place)
I1117 20:11:01.241883 1012 net.cpp:150] Setting up rpn_relu/3x3
I1117 20:11:01.241888 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.241890 1012 net.cpp:165] Memory required for data: 116006912
I1117 20:11:01.241892 1012 layer_factory.hpp:77] Creating layer rpn/output_rpn_relu/3x3_0_split
I1117 20:11:01.241896 1012 net.cpp:106] Creating Layer rpn/output_rpn_relu/3x3_0_split
I1117 20:11:01.241899 1012 net.cpp:454] rpn/output_rpn_relu/3x3_0_split <- rpn/output
I1117 20:11:01.241901 1012 net.cpp:411] rpn/output_rpn_relu/3x3_0_split -> rpn/output_rpn_relu/3x3_0_split_0
I1117 20:11:01.241905 1012 net.cpp:411] rpn/output_rpn_relu/3x3_0_split -> rpn/output_rpn_relu/3x3_0_split_1
I1117 20:11:01.241945 1012 net.cpp:150] Setting up rpn/output_rpn_relu/3x3_0_split
I1117 20:11:01.241961 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.241964 1012 net.cpp:157] Top shape: 1 512 14 14 (100352)
I1117 20:11:01.241966 1012 net.cpp:165] Memory required for data: 116809728
I1117 20:11:01.241967 1012 layer_factory.hpp:77] Creating layer rpn_cls_score
I1117 20:11:01.241972 1012 net.cpp:106] Creating Layer rpn_cls_score
I1117 20:11:01.241976 1012 net.cpp:454] rpn_cls_score <- rpn/output_rpn_relu/3x3_0_split_0
I1117 20:11:01.241977 1012 net.cpp:411] rpn_cls_score -> rpn_cls_score
I1117 20:11:01.242676 1012 net.cpp:150] Setting up rpn_cls_score
I1117 20:11:01.242684 1012 net.cpp:157] Top shape: 1 18 14 14 (3528)
I1117 20:11:01.242686 1012 net.cpp:165] Memory required for data: 116823840
I1117 20:11:01.242691 1012 layer_factory.hpp:77] Creating layer rpn_bbox_pred
I1117 20:11:01.242697 1012 net.cpp:106] Creating Layer rpn_bbox_pred
I1117 20:11:01.242698 1012 net.cpp:454] rpn_bbox_pred <- rpn/output_rpn_relu/3x3_0_split_1
I1117 20:11:01.242702 1012 net.cpp:411] rpn_bbox_pred -> rpn_bbox_pred
I1117 20:11:01.243345 1012 net.cpp:150] Setting up rpn_bbox_pred
I1117 20:11:01.243353 1012 net.cpp:157] Top shape: 1 36 14 14 (7056)
I1117 20:11:01.243355 1012 net.cpp:165] Memory required for data: 116852064
I1117 20:11:01.243360 1012 layer_factory.hpp:77] Creating layer rpn_cls_score_reshape
I1117 20:11:01.243369 1012 net.cpp:106] Creating Layer rpn_cls_score_reshape
I1117 20:11:01.243371 1012 net.cpp:454] rpn_cls_score_reshape <- rpn_cls_score
I1117 20:11:01.243376 1012 net.cpp:411] rpn_cls_score_reshape -> rpn_cls_score_reshape
I1117 20:11:01.243392 1012 net.cpp:150] Setting up rpn_cls_score_reshape
I1117 20:11:01.243396 1012 net.cpp:157] Top shape: 1 2 126 14 (3528)
I1117 20:11:01.243398 1012 net.cpp:165] Memory required for data: 116866176
I1117 20:11:01.243401 1012 layer_factory.hpp:77] Creating layer rpn_cls_prob
I1117 20:11:01.243403 1012 net.cpp:106] Creating Layer rpn_cls_prob
I1117 20:11:01.243405 1012 net.cpp:454] rpn_cls_prob <- rpn_cls_score_reshape
I1117 20:11:01.243409 1012 net.cpp:411] rpn_cls_prob -> rpn_cls_prob
I1117 20:11:01.243543 1012 net.cpp:150] Setting up rpn_cls_prob
I1117 20:11:01.243548 1012 net.cpp:157] Top shape: 1 2 126 14 (3528)
I1117 20:11:01.243551 1012 net.cpp:165] Memory required for data: 116880288
I1117 20:11:01.243552 1012 layer_factory.hpp:77] Creating layer rpn_cls_prob_reshape
I1117 20:11:01.243556 1012 net.cpp:106] Creating Layer rpn_cls_prob_reshape
I1117 20:11:01.243558 1012 net.cpp:454] rpn_cls_prob_reshape <- rpn_cls_prob
I1117 20:11:01.243562 1012 net.cpp:411] rpn_cls_prob_reshape -> rpn_cls_prob_reshape
I1117 20:11:01.243577 1012 net.cpp:150] Setting up rpn_cls_prob_reshape
I1117 20:11:01.243580 1012 net.cpp:157] Top shape: 1 18 14 14 (3528)
I1117 20:11:01.243583 1012 net.cpp:165] Memory required for data: 116894400
I1117 20:11:01.243585 1012 layer_factory.hpp:77] Creating layer proposal
I1117 20:11:01.253264 1012 net.cpp:106] Creating Layer proposal
I1117 20:11:01.253273 1012 net.cpp:454] proposal <- rpn_cls_prob_reshape
I1117 20:11:01.253276 1012 net.cpp:454] proposal <- rpn_bbox_pred
I1117 20:11:01.253279 1012 net.cpp:454] proposal <- im_info
I1117 20:11:01.253283 1012 net.cpp:411] proposal -> rois
I1117 20:11:01.253809 1012 net.cpp:150] Setting up proposal
I1117 20:11:01.253818 1012 net.cpp:157] Top shape: 1 5 (5)
I1117 20:11:01.253820 1012 net.cpp:165] Memory required for data: 116894420
I1117 20:11:01.253823 1012 layer_factory.hpp:77] Creating layer roi_pool5
I1117 20:11:01.253828 1012 net.cpp:106] Creating Layer roi_pool5
I1117 20:11:01.253829 1012 net.cpp:454] roi_pool5 <- conv5_3_relu5_3_0_split_1
I1117 20:11:01.253832 1012 net.cpp:454] roi_pool5 <- rois
I1117 20:11:01.253835 1012 net.cpp:411] roi_pool5 -> pool5
I1117 20:11:01.253840 1012 roi_pooling_layer.cpp:30] Spatial scale: 0.0625
I1117 20:11:01.253870 1012 net.cpp:150] Setting up roi_pool5
I1117 20:11:01.253875 1012 net.cpp:157] Top shape: 1 512 7 7 (25088)
I1117 20:11:01.253877 1012 net.cpp:165] Memory required for data: 116994772
I1117 20:11:01.253878 1012 layer_factory.hpp:77] Creating layer fc6
I1117 20:11:01.253886 1012 net.cpp:106] Creating Layer fc6
I1117 20:11:01.253890 1012 net.cpp:454] fc6 <- pool5
I1117 20:11:01.253891 1012 net.cpp:411] fc6 -> fc6
I1117 20:11:01.372174 1012 net.cpp:150] Setting up fc6
I1117 20:11:01.372191 1012 net.cpp:157] Top shape: 1 4096 (4096)
I1117 20:11:01.372194 1012 net.cpp:165] Memory required for data: 117011156
I1117 20:11:01.372220 1012 layer_factory.hpp:77] Creating layer relu6
I1117 20:11:01.372227 1012 net.cpp:106] Creating Layer relu6
I1117 20:11:01.372231 1012 net.cpp:454] relu6 <- fc6
I1117 20:11:01.372234 1012 net.cpp:397] relu6 -> fc6 (in-place)
I1117 20:11:01.372676 1012 net.cpp:150] Setting up relu6
I1117 20:11:01.372684 1012 net.cpp:157] Top shape: 1 4096 (4096)
I1117 20:11:01.372685 1012 net.cpp:165] Memory required for data: 117027540
I1117 20:11:01.372689 1012 layer_factory.hpp:77] Creating layer fc7
I1117 20:11:01.372692 1012 net.cpp:106] Creating Layer fc7
I1117 20:11:01.372709 1012 net.cpp:454] fc7 <- fc6
I1117 20:11:01.372712 1012 net.cpp:411] fc7 -> fc7
I1117 20:11:01.393120 1012 net.cpp:150] Setting up fc7
I1117 20:11:01.393136 1012 net.cpp:157] Top shape: 1 4096 (4096)
I1117 20:11:01.393138 1012 net.cpp:165] Memory required for data: 117043924
I1117 20:11:01.393159 1012 layer_factory.hpp:77] Creating layer relu7
I1117 20:11:01.393167 1012 net.cpp:106] Creating Layer relu7
I1117 20:11:01.393170 1012 net.cpp:454] relu7 <- fc7
I1117 20:11:01.393173 1012 net.cpp:397] relu7 -> fc7 (in-place)
I1117 20:11:01.393369 1012 net.cpp:150] Setting up relu7
I1117 20:11:01.393374 1012 net.cpp:157] Top shape: 1 4096 (4096)
I1117 20:11:01.393376 1012 net.cpp:165] Memory required for data: 117060308
I1117 20:11:01.393378 1012 layer_factory.hpp:77] Creating layer fc7_relu7_0_split
I1117 20:11:01.393383 1012 net.cpp:106] Creating Layer fc7_relu7_0_split
I1117 20:11:01.393384 1012 net.cpp:454] fc7_relu7_0_split <- fc7
I1117 20:11:01.393401 1012 net.cpp:411] fc7_relu7_0_split -> fc7_relu7_0_split_0
I1117 20:11:01.393405 1012 net.cpp:411] fc7_relu7_0_split -> fc7_relu7_0_split_1
I1117 20:11:01.393458 1012 net.cpp:150] Setting up fc7_relu7_0_split
I1117 20:11:01.393461 1012 net.cpp:157] Top shape: 1 4096 (4096)
I1117 20:11:01.393465 1012 net.cpp:157] Top shape: 1 4096 (4096)
I1117 20:11:01.393466 1012 net.cpp:165] Memory required for data: 117093076
I1117 20:11:01.393467 1012 layer_factory.hpp:77] Creating layer cls_score
I1117 20:11:01.393471 1012 net.cpp:106] Creating Layer cls_score
I1117 20:11:01.393473 1012 net.cpp:454] cls_score <- fc7_relu7_0_split_0
I1117 20:11:01.393491 1012 net.cpp:411] cls_score -> cls_score
I1117 20:11:01.394181 1012 net.cpp:150] Setting up cls_score
I1117 20:11:01.394187 1012 net.cpp:157] Top shape: 1 21 (21)
I1117 20:11:01.394189 1012 net.cpp:165] Memory required for data: 117093160
I1117 20:11:01.394193 1012 layer_factory.hpp:77] Creating layer bbox_pred
I1117 20:11:01.394196 1012 net.cpp:106] Creating Layer bbox_pred
I1117 20:11:01.394214 1012 net.cpp:454] bbox_pred <- fc7_relu7_0_split_1
I1117 20:11:01.394218 1012 net.cpp:411] bbox_pred -> bbox_pred
I1117 20:11:01.394469 1012 net.cpp:150] Setting up bbox_pred
I1117 20:11:01.394474 1012 net.cpp:157] Top shape: 1 84 (84)
I1117 20:11:01.394475 1012 net.cpp:165] Memory required for data: 117093496
I1117 20:11:01.394479 1012 layer_factory.hpp:77] Creating layer cls_prob
I1117 20:11:01.394482 1012 net.cpp:106] Creating Layer cls_prob
I1117 20:11:01.394484 1012 net.cpp:454] cls_prob <- cls_score
I1117 20:11:01.394501 1012 net.cpp:411] cls_prob -> cls_prob
I1117 20:11:01.394683 1012 net.cpp:150] Setting up cls_prob
I1117 20:11:01.394688 1012 net.cpp:157] Top shape: 1 21 (21)
I1117 20:11:01.394690 1012 net.cpp:165] Memory required for data: 117093580
I1117 20:11:01.394692 1012 net.cpp:228] cls_prob does not need backward computation.
I1117 20:11:01.394695 1012 net.cpp:228] bbox_pred does not need backward computation.
I1117 20:11:01.394697 1012 net.cpp:228] cls_score does not need backward computation.
I1117 20:11:01.394700 1012 net.cpp:228] fc7_relu7_0_split does not need backward computation.
I1117 20:11:01.394716 1012 net.cpp:228] relu7 does not need backward computation.
I1117 20:11:01.394717 1012 net.cpp:228] fc7 does not need backward computation.
I1117 20:11:01.394721 1012 net.cpp:228] relu6 does not need backward computation.
I1117 20:11:01.394722 1012 net.cpp:228] fc6 does not need backward computation.
I1117 20:11:01.394726 1012 net.cpp:228] roi_pool5 does not need backward computation.
I1117 20:11:01.394728 1012 net.cpp:228] proposal does not need backward computation.
I1117 20:11:01.394732 1012 net.cpp:228] rpn_cls_prob_reshape does not need backward computation.
I1117 20:11:01.394749 1012 net.cpp:228] rpn_cls_prob does not need backward computation.
I1117 20:11:01.394752 1012 net.cpp:228] rpn_cls_score_reshape does not need backward computation.
I1117 20:11:01.394755 1012 net.cpp:228] rpn_bbox_pred does not need backward computation.
I1117 20:11:01.394773 1012 net.cpp:228] rpn_cls_score does not need backward computation.
I1117 20:11:01.394775 1012 net.cpp:228] rpn/output_rpn_relu/3x3_0_split does not need backward computation.
I1117 20:11:01.394778 1012 net.cpp:228] rpn_relu/3x3 does not need backward computation.
I1117 20:11:01.394793 1012 net.cpp:228] rpn_conv/3x3 does not need backward computation.
I1117 20:11:01.394796 1012 net.cpp:228] conv5_3_relu5_3_0_split does not need backward computation.
I1117 20:11:01.394798 1012 net.cpp:228] relu5_3 does not need backward computation.
I1117 20:11:01.394801 1012 net.cpp:228] conv5_3 does not need backward computation.
I1117 20:11:01.394804 1012 net.cpp:228] relu5_2 does not need backward computation.
I1117 20:11:01.394806 1012 net.cpp:228] conv5_2 does not need backward computation.
I1117 20:11:01.394824 1012 net.cpp:228] relu5_1 does not need backward computation.
I1117 20:11:01.394827 1012 net.cpp:228] conv5_1 does not need backward computation.
I1117 20:11:01.394829 1012 net.cpp:228] pool4 does not need backward computation.
I1117 20:11:01.394846 1012 net.cpp:228] relu4_3 does not need backward computation.
I1117 20:11:01.394850 1012 net.cpp:228] conv4_3 does not need backward computation.
I1117 20:11:01.394852 1012 net.cpp:228] relu4_2 does not need backward computation.
I1117 20:11:01.394855 1012 net.cpp:228] conv4_2 does not need backward computation.
I1117 20:11:01.394858 1012 net.cpp:228] relu4_1 does not need backward computation.
I1117 20:11:01.394873 1012 net.cpp:228] conv4_1 does not need backward computation.
I1117 20:11:01.394876 1012 net.cpp:228] pool3 does not need backward computation.
I1117 20:11:01.394879 1012 net.cpp:228] relu3_3 does not need backward computation.
I1117 20:11:01.394882 1012 net.cpp:228] conv3_3 does not need backward computation.
I1117 20:11:01.394886 1012 net.cpp:228] relu3_2 does not need backward computation.
I1117 20:11:01.394887 1012 net.cpp:228] conv3_2 does not need backward computation.
I1117 20:11:01.394891 1012 net.cpp:228] relu3_1 does not need backward computation.
I1117 20:11:01.394893 1012 net.cpp:228] conv3_1 does not need backward computation.
I1117 20:11:01.394896 1012 net.cpp:228] pool2 does not need backward computation.
I1117 20:11:01.394898 1012 net.cpp:228] relu2_2 does not need backward computation.
I1117 20:11:01.394901 1012 net.cpp:228] conv2_2 does not need backward computation.
I1117 20:11:01.394904 1012 net.cpp:228] relu2_1 does not need backward computation.
I1117 20:11:01.394906 1012 net.cpp:228] conv2_1 does not need backward computation.
I1117 20:11:01.394909 1012 net.cpp:228] pool1 does not need backward computation.
I1117 20:11:01.394912 1012 net.cpp:228] relu1_2 does not need backward computation.
I1117 20:11:01.394914 1012 net.cpp:228] conv1_2 does not need backward computation.
I1117 20:11:01.394917 1012 net.cpp:228] relu1_1 does not need backward computation.
I1117 20:11:01.394919 1012 net.cpp:228] conv1_1 does not need backward computation.
I1117 20:11:01.394922 1012 net.cpp:270] This network produces output bbox_pred
I1117 20:11:01.394925 1012 net.cpp:270] This network produces output cls_prob
I1117 20:11:01.394944 1012 net.cpp:283] Network initialization done.
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:505] Reading dangerously large protocol message. If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons. To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING google/protobuf/io/coded_stream.cc:78] The total number of bytes read was 548317115
I1117 20:11:01.871199 1012 net.cpp:816] Ignoring source layer data
I1117 20:11:01.927953 1012 net.cpp:816] Ignoring source layer drop6
I1117 20:11:01.935667 1012 net.cpp:816] Ignoring source layer drop7
I1117 20:11:01.935674 1012 net.cpp:816] Ignoring source layer fc7_drop7_0_split
I1117 20:11:01.935938 1012 net.cpp:816] Ignoring source layer loss_cls
I1117 20:11:01.935943 1012 net.cpp:816] Ignoring source layer loss_bbox
I1117 20:11:01.937173 1012 net.cpp:816] Ignoring source layer silence_rpn_cls_score
I1117 20:11:01.937180 1012 net.cpp:816] Ignoring source layer silence_rpn_bbox_pred

Loaded network /media/xinleipan/DC1A-EA60/git14/py-faster-rcnn/data/faster_rcnn_models/VGG16_faster_rcnn_final.caffemodel

Demo for data/demo/000456.jpg
Detection took 0.044s for 300 object proposals

Demo for data/demo/000542.jpg
Detection took 0.041s for 259 object proposals

Demo for data/demo/001150.jpg
Detection took 0.040s for 223 object proposals

Demo for data/demo/001763.jpg
Detection took 0.038s for 201 object proposals

Demo for data/demo/004545.jpg
Detection took 0.029s for 172 object proposals

Obviously, it fails to detect the object in the image. When I turn off GPU with "--cpu" it works. I'm not sure what is the problem?

xinleipan on 18 Nov 2016

re xinlei:
Have you linked cudnn to /usr/local/cuda?
what is your Makefile.config?

max-ng on 18 Nov 2016

@neoscode I'm not sure ... I actually commented use CUDNN and tried again and it works pretty well....

So I guess the reason is that I didn't install cudnn correctly. I followed the installation on tensorflow insallation website:
https://www.tensorflow.org/versions/r0.11/get_started/os_setup.html#optional-install-cuda-gpus-on-linux
and did this thing:
tar xvzf cudnn-8.0-linux-x64-v5.1-ga.tgz
sudo cp cuda/include/cudnn.h /usr/local/cuda/include
sudo cp cuda/lib64/libcudnn* /usr/local/cuda/lib64
sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*

maybe it is not correct way to install cudnn.

xinleipan on 19 Nov 2016

👍1

It seems like you are also Chinese and interested in machine learning. Lets make a friend. My email is [email protected]

max-ng on 19 Nov 2016

First step: Install CUDA 8.0 and CUDNN 5.0 for CUDA 8.0: https://www.pugetsystems.com/labs/hpc/Install-Ubuntu-16-04-or-14-04-and-CUDA-8-and-7-5-for-NVIDIA-Pascal-GPU-825/
Download Faster-RCNN from this repo. Before you make caffe, do these steps first: