shaoqingren / faster_rcnn Goto Github PK

I've complied your Caffe branch successfully, make test works OK as well, but running make runtest, leads to an error regarding an
"Unknown V1LayerParameter layer type:40".

There was also an issue with compiling matcaffe, giving an error of "gpu/mxGPUArray.h" not found in cafe_.cpp

Any possible insights will be greatly appreciated! Thanks!

In the m file roidb_from_voc.m, the function attach_proposals(~) has some codes like this:
///////////////////////////////////////////////////
function rec = attach_proposals(voc_rec, boxes, class_to_id, exclude_difficult_samples, flip)
……
if isfield(voc_rec, 'objects')
……
else
gt_boxes = [];
all_boxes = boxes;
gt_classes = [];
num_gt_boxes = 0;
end
//////////////////////////////////////////////////////
So, when our datasets has a image with no objects, gt_boxes = [];

so in the script_faster_rcnn_VOC0712_ZF.m at stage one proposal , this line:
model.stage1_rpn = Faster_RCNN_Train.do_proposal_train(conf_proposal, dataset, model.stage1_rpn, opts.do_val);
will input dataset, and dataset.roidb_train(p).rois(q).boxes = [];

so int the file proposal_train.m with this line:
[image_roidb_train, bbox_means, bbox_stds]...
= proposal_prepare_image_roidb(conf, opts.imdb_train, opts.roidb_train);
will input opts.roidb_train, and opts.roidb_train(p).rois(q) = [];

thus in the file proposal_prepare_image_roidb.m with this function will failed:
function scaled_rois = scale_rois(rois, im_size, im_scale)
im_size_scaled = round(im_size * im_scale);
scale = (im_size_scaled - 1) ./ (im_size - 1);
scaled_rois = bsxfun(@times, rois-1, [scale(2), scale(1), scale(2), scale(1)]) + 1;
end
Because rois = [], and bsxfun has two input array with different dimension, errors：
//////////////////////////////////////////////////////////////////////////////////
Error using proposal_prepare_image_roidb>scale_rois (line
112)
Non-singleton dimensions of the two input arrays must match
each other.

Error in
proposal_prepare_image_roidb>@(x,y)compute_targets(conf,scale_rois(gt_rois,im_size,y),gt_labels,x,image_roidb_cell{i},y)
(line 62)
bbox_targets{i} = cellfun(@(x, y) ...

Error in proposal_prepare_image_roidb>(parfor body) (line 62)
bbox_targets{i} = cellfun(@(x, y) ...

Error in
proposal_prepare_image_roidb>append_bbox_regression_targets
(line 53)
parfor i = 1:num_images

Error in proposal_prepare_image_roidb (line 43)
[image_roidb, bbox_means, bbox_stds] =
append_bbox_regression_targets(conf, image_roidb,
bbox_means, bbox_stds);

Error in proposal_train (line 76)
[image_roidb_train, bbox_means, bbox_stds]...

Error in Faster_RCNN_Train.do_proposal_train (line 7)
model_stage.output_model_file = proposal_train(conf,
dataset.imdb_train, dataset.roidb_train, ...

Error in script_faster_rcnn_pku (line 46)
model.stage1_rpn =
Faster_RCNN_Train.do_proposal_train(conf_proposal, dataset,
model.stage1_rpn, opts.do_val);

Caused by:
Error using bsxfun
Non-singleton dimensions of the two input arrays must
match each other.
///////////////////////////////////////////////////////////////////////////////////////////////

So that means the previous if……else……end code block is wrong ???
and how to deal with the problem when our image training dataset has no objects in a image ?

I know the mex compatibility for faster_rcnn is vs2013 ,windows8 and cuda 6.5.
However is there anyway I can make it run under cuda 7.5?
Thank you!

Dear ShaoqingRen:
I am trying to use the code to train a ZF model by running 'script_fast_rcnn_VOC2007_ZF". It
failed though I have downloaded the required model and voc2007 dataset.
Then "with_selective_search" option was disabled by set "dataset.imdb_test.roidb_func(dataset.imdb_test, 'with_selective_search', false);".It failed again
because a warning was outputed as "Warning: fast_rcnn_prepare_image_roidb: filter out 10022
images, which contains zero valid samples".
Does selective_search boxes must be presented while training? What can I do to make train going down?

hello ! Thank you for making the code available, much appreciated. I was formerly working with caffe and selective search, instead of using rcnn I had written my own scripts, I have bumped into every problem with computational cost of localization so I want to experiment right away

Anyhow, I went over the code and I thought a mini guide for training a whole new dataset from scratch would be much appreciated as I have a few questions in my head. Is it possible to start training without finetuning from a pretrained model? (I do have a one class (pos or neg/object or b.g. (not object)) problem which is not an object in any of the classes of VOC or imagenet, so I believe I need a separate model for objectness trained on my dataset) Do I train a network originally defined as in caffe paper using annotation boxes as positives? Do I need to compute image mean myself? If so, is the image mean of the whole frame or the annotation box? I am assuming it should be in caffe channels order? Since I keep getting the libprotobuf error it is making it difficult to trace in debug mode ( I can't run any function twice on matlab)

I would really appreciate the tips to train on a whole new dataset (which I can arrange in voc format ) if anyone has already attempted to do so.Thanks !

After the impressive results of replacing VGG16 model with the recent ResNet-101 in the new MSRA paper, I think we would all love to be able to experiment with the new model.

@ShaoqingRen Are you planning on making the new model available here? Thanks in advance! And thanks for making faster r-cnn really awesome!

Dear Shaoqing,

I am wondering what is the acutal meanning for scale_train in your dropout layer. I cannot find such parameter in original Caffe project.

Because i encountered an unknown error when runing your caffe_.mexw64, i have to use my own, but it is not compatible to your prototxt file.

What do you think is the best solution to my issue?

Thanks,

I tried to run 'script_faster_rcnn_VOC2007_VGG16.m' on Ubuntu14.04 but I got this error

F1031 20:05:16.437945 12590 io.cpp:34] Check failed: fd != -1 (-1 vs. -1) File not found: .\models\rpn_prototxts\vgg_16layers_conv3_1\train_val.prototxt

Of course, I've followed instructions and found the file exists in the same place.
In addition, I tried
addpath(genpath(fullfile(curdir, 'models')));
at startup.m.
But the error still remains.
Is there anyone who suffers same error?

Thank you in advance.

@ShaoqingRen It is very exciting to know that your MSRA team have broken all the records on ImageNet classification, ImageNet detection, ImageNet localization, COCO detection, COCO segmentation and PASCAL VOC detection with large margins. Your achievements mark very significant milestones in the related areas. When and where will the ResNet paper [1] be released? Thanks a lot!

Sorry about that this question is not directly related to the project. I just can't wait to read the paper.

[1] "Deep Residual Learning for Image Recognition", Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun. Tech Report 2015.

i have my own pictures,and want figure it like voc2007,have annoation,imagesets...,eg.did you have any util like this?
is any qq group i can join?

In the paper，said input Map is of any size. why the prototxt rpn training have the data layer with dim 1_3_224*224? 3Q!

When I run script_faster_rcnn_demo for the first time after starting Matlab, things work fine. But if I re-run the same script after the first run, I get the following error

fast_rcnn startup done
GPU 1: free memory 11945885696
GPU 2: free memory 813449216
Use GPU 1
[libprotobuf ERROR google/protobuf/descriptor_database.cc:57] File already exists in database: caffe.proto
[libprotobuf FATAL google/protobuf/descriptor.cc:954] CHECK failed: generated_database_->Add(encoded_file_descriptor, size):
Caught "std::exception" Exception message is:
CHECK failed: generated_database_->Add(encoded_file_descriptor, size):

I tried to run 'script_faster_rcnn_demo.m' but I got this error

error: caffe.set_device (line 9)
caffe_('set_device', device_id);

Matlab said 'caffe_' is not defined.

Startup.m added 'external/caffe/matlab' without trouble and
I checked 'external/caffe/matlab/+caffe/private' folder and there is 'caffe_.mexa64' but it seems Matlab cannot use this file.

However, 'classification_demo.m' in 'external/caffe/matlab/demo' works.

Is there anyone who suffers same error?

i success train by my datasets ,now i want use it in mobile unit or produce environment,is matlab easy transplant?

Hi all:
I would like to train faster-rcnn on my own dataset, but i do not know what data files should be prapared？? i.e. the images, annotations? Any reference blogs or tutorials will be highly appreciated. Thanks

Hi,

I am trying to retrain ZF or VGG model using only two categories from VOC2007 (dog and cat). I have cleaned Annotations and ImageSets directories from other categories and left only necessary images in JPEGImages directory. I have changed outputs of layers in particular .prototxt files from 84 = 4(20 categories + 1 background) to 12 = 4*(2 categories + 1 background).

The first step of training (Train RPN with conv layers tuned; compute RPN results on the train/test sets) is performed without any problems. However when the second step starts training fails with error in math_functions.cu Check failed: status == CUBLAS_STATUS_SUCCESS (13 vs. 0) CUBLAS_STATUS_EXECUTION_FAILED. I have found out that the problem is caused in either caffe_gpu_axpy() or caffe_gpu_scal() only when parameter N has larger value. I tried to decrease values of fast rcnn settings in fast_rcnn_config.m, but the same error still persists.

Any idea what am I doing wrong? :)

I am using Ubuntu 14.04.3 LTS, Matlab R2012a and Tesla K40m.

Hi...

I'm on Ubuntu Linux 14.4, and i'm trying to run the example.

I get to the "Run experiments/script_faster_rcnn_demo.m to test a single demo image." point, but i get the error:

Error using cd
Cannot CD to
/home/kasper/masterthesis/faster_rcnn/experiments/external/caffe/matlab/caffe_faster_rcnn (Name
is nonexistent or not a directory).

I have built caffe with make like I've done a hundred times before with regular caffe, but the caffe_faster_cnn part does not seem to be built.

i followed the instruction of https://github.com/ShaoqingRen/caffe/tree/faster-R-CNN to create my mex-File( win8.1, matlab 2014a , vs2013, cuda 7.5) but when i tried to test the faster-rcnn demo i had this problem

GPU 1: free memory 2059571200
Use GPU 1
Invalid MEX-file 'C:\Users\Public\faster_rcnn-master\external\caffe\matlab\caffe_faster_rcnn\caffe_.mexw64': The specified
module could not be found.

Error in caffe.set_device (line 9)
caffe_('set_device', device_id);

Error in active_caffe_mex (line 26)
caffe.set_device(gpu_id-1);

Error in script_faster_rcnn_demo (line 10)
active_caffe_mex(opts.gpu_id, opts.caffe_version);

In the paper，said input Map is of any size. why the prototxt rpn training have the data layer with dim 1_3_224*224? 3Q!

Hi shaoqing,

I trained the ZF model on VOC2007 dataset by runing script_faster_rcnn_VOC2007_ZF.m and get an mAP ~55.6, this is lower than the ~59.4 mAP in you training log file (fast_rcnn_VOC2007_ZF/voc_2007_test/log/test_20150820_194058.txt).

Is there any difference between the 2 training processes? Here is my output

Results:
   60.7088
   67.7598
   53.3626
   37.1256
   24.7806
   60.3213
   71.7281
   68.8887
   30.3973
   62.4456
   56.2178
   63.1661
   73.3527
   64.2983
   61.4177
   27.1311
   55.2103
   51.5869
   68.5139
   53.5203

   55.5967

Cleared 0 solvers and 2 stand-alone nets
please modify detection_test.prototxt file for sharing conv layers with proposal model (delete layers until relu5

I am using Matlab R2013b, and I can run testing perfectly. However when I do training (e.g., VOC2007 on ZF net), the process gets stuck at

caffe_solver = caffe.Solver(opts.solver_def_file);

in 'proposal_train.m'.

It seems that it's just an initialization of Caffe.

Test on 2 machine and get the same problem. Does anyone meet with this, too?

I now think in four categories (car, truck, bus, motorcycle), for my own project
VOC categories changed, whether i can aslo use zf.caffemodel or vgg16.caffemodel to initialize the RPN and Fast_rcnn?

if not, I should be how to adjust or create a new

I need to detect cars or pedestrians from a UAV, such as in the car parking lot. However, there exist a planar rotation in the aerial image. Are there some techniques to handle such case?

when i run script_faster_rcnn_demo.m in matlab , error occured like that:
Invalid MEX-file
'flles...\caffe_.mexw64':
找不到指定的模块. but the caffe mex file is there, i download the caffe mex from baidu yunpan following the dirctory from readme file , i didnt know what is going wrong .

Hi,

I got a problem with compiling caffe. When I run "make matcaffe" I got an error, the outputs are as below:
MEX matlab/+caffe/private/caffe_.cpp

Warning: You are using gcc version "4.8.2-19ubuntu1)". The version
currently supported with MEX is "4.4.6".
For a list of currently supported compilers see:
http://www.mathworks.com/support/compilers/current_release/

matlab/+caffe/private/caffe_.cpp:19:28: fatal error: gpu/mxGPUArray.h: No such file or directory
#include "gpu/mxGPUArray.h"
^
compilation terminated.

mex: compile of ' "matlab/+caffe/private/caffe_.cpp"' failed.

make: *** [matlab/+caffe/private/caffe_.mexa64] Error 1

I am using matlab 2012a on my server. And I have search about this problem while I got answer that there is a mxGPUArray.h file in \MATLAB\R2014a\toolbox\distcomp\gpu\extern\include\gpu, however 2012a version seems do not have this file. Because the server is shared by several labmates, I am afraid reinstall the matlab to another version will affect others' program. So is there anyway to solver this problem without chaging the matlab version?

Thanks.

I followed the installation steps and also downloaded the models but when I run the demo script_faster_rcnn_demo.m , I get the error:

fast_rcnn startup done
GPU 1: free memory 5307551744
GPU 2: free memory 6312861696
Use GPU 2

Reference to non-existent field 'bottom_id_vecs'.

Error in caffe.Net (line 74)
self.attributes.bottom_id_vecs = cellfun(@(x) x+1, self.attributes.bottom_id_vecs, 'UniformOutput', false);

Error in caffe.get_net (line 28)
net = caffe.Net(hNet);

Error in caffe.Net (line 33)
self = caffe.get_net(varargin{:});

Error in script_faster_rcnn_demo (line 33)
rpn_net = caffe.Net(proposal_detection_model.proposal_net_def, 'test');

Hi,
I have trained model with the file script_faster_rcnn_VOC2007_ZF.m already by commenting line 55 opts.mAP = Faster_RCNN_Train.do_fast_rcnn_test(conf_fast_rcnn, model.stage1_fast_rcnn, dataset.imdb_test, dataset.roidb_test);
and line 78 opts.final_mAP = Faster_RCNN_Train.do_fast_rcnn_test(conf_fast_rcnn, model.stage2_fast_rcnn, dataset.imdb_test, dataset.roidb_test);

If i didn't do that, error comes as follows:

Error using fprintf,
Invalid file identifier, use open to generate a file identifier.

Error in imdb_eval_voc(line 63)
fprintf(fid, '%s %f %.3f %.3f %.3f %.3f\n', image_ids{i}, bbox(j,end), bbox(j,1:4));

Error in fast_rcnn_test(line 171)
res(model_ind) = imdb.eval_func(cls, aboxes{model_ind}, imdb, opts.cache_name, opts.suffix);

Error in Faster_Rcnn_Train.do_fast_rcnn_test(line 6)
mAP = fast_rcnn_test(conf, imdb, roidb, ...
'net_def_file', model_stage.test_net_def_file, ...
'net_file', model_stage.output_model_file, ...
'cache_name', model_stage.cache_name, ...
'ignore_cache', ignore_cache);

Error in script_faster_rcnn_VOC2007_ZF(line 55)
opts.mAP = Faster_RCNN_Train.do_fast_rcnn_test(conf_fast_rcnn, model.stage1_fast_rcnn, dataset.imdb_test, dataset.roidb_test);

Thanks.

I can successfully run script_faster_rcnn_demo.m with following configs:
opts.use_gpu = false;
model_dir = fullfile(pwd, 'output', 'faster_rcnn_final', 'faster_rcnn_VOC0712_vgg_16layers'); %% VGG-16

opts.use_gpu = false;
model_dir = fullfile(pwd, 'output', 'faster_rcnn_final', 'faster_rcnn_VOC0712_ZF'); %% ZF

opts.use_gpu = true;
model_dir = fullfile(pwd, 'output', 'faster_rcnn_final', 'faster_rcnn_VOC0712_ZF'); %% ZF

But when I run it with this config:
opts.use_gpu = true;
model_dir = fullfile(pwd, 'output', 'faster_rcnn_final', 'faster_rcnn_VOC0712_vgg_16layers'); %% VGG-16
My Matlab software crashed and shutdown with nothing tips, What's the problem?
PS: CUDA version is 7.5, Windows OS, Discrete Graphics is Quadro K620

Invalid MEX-file '...\faster_rcnn-master\external\caffe\matlab\caffe_faster_rcnn\caffe_.mexw64': The specified procedure could not be found.

Error in caffe.get_net (line 27)
hNet = caffe_('get_net', model_file, phase_name);

Error in caffe.Net (line 33)
self = caffe.get_net(varargin{:});

Error in script_faster_rcnn_demo (line 33)
rpn_net = caffe.Net(proposal_detection_model.proposal_net_def, 'test');

Hi, I am unable to find solver and training prototxt files required for training rpn and fast rcnn model.

I got out of memory error in the middle of Faster_RCNN_Train.do_proposal_test, is that out of gpu memory? is there any batch size I can set to reduce the memory.

and I already trained stage 1 RPN, is there anyway I can resume training like caffe does?

experiments: test (voc_2007_trainval) 7559/10022 time: 0.378s
experiments: test (voc_2007_trainval) 7560/10022 time: 0.339s
experiments: test (voc_2007_trainval) 7561/10022 time: 0.307s
experiments: test (voc_2007_trainval) 7562/10022 time: 0.467s
experiments: test (voc_2007_trainval) 7563/10022 {�Error using imresizemex
Out of memory. Type HELP MEMORY for your options.

Error in imresize>resizeAlongDim (line 762)
out = imresizemex(in, weights', indices', dim);

Error in imresize (line 169)
B = resizeAlongDim(B, dim, weights{dim}, indices{dim});

Error in prep_im_for_blob (line 14)
im = imresize(im, target_size, 'bilinear', 'antialiasing', false);

Error in proposal_im_detect>get_image_blob (line 74)
[blob, im_scales] = prep_im_for_blob(im, conf.image_means, conf.test_scales,
conf.test_max_size);

Error in proposal_im_detect (line 10)
[im_blob, im_scales] = get_image_blob(conf, im);

Error in proposal_test (line 80)
[boxes, scores, abox_deltas{i}, aanchors{i}, ascores{i}] =
proposal_im_detect(conf, caffe_net, im);

Error in Faster_RCNN_Train.do_proposal_test (line 2)
aboxes = proposal_test(conf, imdb, ...

Error in
script_faster_rcnn_VOC2007_ZF>@(x,y)Faster_RCNN_Train.do_proposal_test(conf_proposal,model.stage1_rpn,x,y)
(line 47)
dataset.roidb_train = cellfun(@(x, y)
Faster_RCNN_Train.do_proposal_test(conf_proposal, model.stage1_rpn, x, y),
dataset.imdb_train, dataset.roidb_train, 'UniformOutput', false);

Error in script_faster_rcnn_VOC2007_ZF (line 47)
dataset.roidb_train = cellfun(@(x, y)
Faster_RCNN_Train.do_proposal_test(conf_proposal, model.stage1_rpn, x, y),
dataset.imdb_train, dataset.roidb_train, 'UniformOutput', false);

opts.per_nms_topN = 6000;
opts.nms_overlap_thres = 0.7;
opts.after_nms_topN = 300;
opts.test_scales = 600;
thankyou

Hi, I am trying to train Faster RCNN with ZFNet on ImageNet data. When I get to the "stage one fast rcnn", i could not pass the memory check for lack of memory. I try to minimize the batch_num to 1 and the problem still exist. Is that because Fast RCNN will load all ROI into GPU memory? I change the nms proposal number and it won't help either.
I have a 2G memory GPU and training RPN is fine.
Ray

I wrote the following function to test the final faster_rcnn model on VOC2007, but I get the error below when the line Faster_RCNN_Train.do_fast_rcnn_test runs. I am guessing this is an issue of input dimensions for my data, but I am not sure what needs to be modified. This script was derived from the training and demo scripts.

F1204 10:30:14.428231 17510 inner_product_layer.cpp:61] Check failed: K_ == new_K (25088 vs. 147) Input size incompatible with inner product parameters.
*** Check failure stack trace: ***

function detect_with_map()
close all;
clc;
clear mex;
clear is_valid_handle; % to clear init_key
run(fullfile(fileparts(fileparts(mfilename('fullpath'))), 'startup'));

%% -------------------- CONFIG --------------------
opts.caffe_version          = 'caffe_faster_rcnn';
opts.gpu_id                 = auto_select_gpu;
active_caffe_mex(opts.gpu_id, opts.caffe_version);

 opts.per_nms_topN           = 6000;
 opts.nms_overlap_thres      = 0.70;
 opts.after_nms_topN          =300;
 opts.use_gpu                = true;
% 
opts.test_scales            = 600;
% 

 %% -------------------- INIT_MODEL --------------------
 model_dir                   = fullfile(pwd, 'output', 'faster_rcnn_final', 'faster_rcnn_VOC0712_vgg_16layers'); %% VGG-16
 %model_dir                   = fullfile(pwd, 'output', 'faster_rcnn_final', 'faster_rcnn_VOC0712_ZF'); %% ZF
 proposal_detection_model    = load_proposal_detection_model(model_dir)

 proposal_detection_model.conf_proposal.test_scales = opts.test_scales;
 proposal_detection_model.conf_detection.test_scales = opts.test_scales;
 proposal_detection_model.conf_proposal.use_gpu = opts.use_gpu;
 proposal_detection_model.conf_detection.use_gpu = opts.use_gpu;
 if opts.use_gpu
     proposal_detection_model.conf_proposal.image_means = gpuArray(proposal_detection_model.conf_proposal.image_means);
     proposal_detection_model.conf_detection.image_means = gpuArray(proposal_detection_model.conf_detection.image_means);
 end

%%MAP Detection
devkit                      = './datasets/VOCdevkit2007';
dataset                     = [];
dataset                     = Dataset.voc2007_test(dataset, 'test', false);

% model
model                       = proposal_detection_model;

%% final test params
model.final_test.nms.per_nms_topN               = 6000; % to speed up nms
model.final_test.nms.nms_overlap_thres          = 0.7;
model.final_test.nms.after_nms_topN             = 300;

%% cache base
cache_base_proposal         = 'faster_rcnn_VOC0712_vgg_16layers';
cache_base_fast_rcnn        = '';

%conf
conf_proposal               = model.conf_proposal;
conf_fast_rcnn              = model.conf_detection; 

det_model=[];
det_model.test_net_def_file = model.detection_net_def;
det_model.output_model_file = model.detection_net;
det_model.cache_name = [cache_base_proposal, '_detection'];
det_model.nms = model.final_test.nms;

prop_model=[];
prop_model.test_net_def_file = model.proposal_net_def;
prop_model.output_model_file = model.proposal_net;
prop_model.cache_name = [cache_base_proposal, '_proposal'];
prop_model.nms = model.final_test.nms;


dataset.roidb_test              = Faster_RCNN_Train.do_proposal_test(conf_proposal, prop_model, dataset.imdb_test, dataset.roidb_test);

opts.final_mAP              = Faster_RCNN_Train.do_fast_rcnn_test(conf_fast_rcnn, det_model, dataset.imdb_test, dataset.roidb_test,true)

end

Thanks in advance!

Environment:
CentOS release 6.3
gcc version 4.8.2 (GCC)
Maltab R2013a (8.1.0.604) 64-bit (glnxa64)

When run 'script_faster_rcnn_demo.m', I got this error:

....
I0915 13:13:05.486282 28509 net.cpp:746] Copying source layer fc6
I0915 13:13:05.581735 28509 net.cpp:746] Copying source layer relu6
I0915 13:13:05.581773 28509 net.cpp:746] Copying source layer drop6
I0915 13:13:05.581779 28509 net.cpp:746] Copying source layer fc7
I0915 13:13:05.597371 28509 net.cpp:746] Copying source layer relu7
I0915 13:13:05.597409 28509 net.cpp:746] Copying source layer drop7
I0915 13:13:05.597415 28509 net.cpp:746] Copying source layer fc7_drop7_0_split
I0915 13:13:05.597421 28509 net.cpp:746] Copying source layer cls_score
I0915 13:13:05.597509 28509 net.cpp:746] Copying source layer bbox_pred
I0915 13:13:05.597848 28509 net.cpp:746] Copying source layer cls_prob
Error using imresize
Expected input number 1, A, to be one of these types:

numeric, logical

Instead its type was gpuArray.

Error in imresize>parsePreMethodArgs (line 333)
validateattributes(A, {'numeric', 'logical'}, {'nonsparse', 'nonempty'}, mfilename, 'A', 1);

Error in imresize>parseInputs (line 248)
[params.A, params.map, params.scale, params.output_size] = ...

Error in imresize (line 141)
params = parseInputs(varargin{:});

Error in prep_im_for_blob (line 30)
im = imresize(im, im_scale);

Error in proposal_im_detect>get_image_blob (line 74)
[blob, im_scales] = prep_im_for_blob(im, conf.image_means, conf.test_scales, conf.test_max_size);

Error in proposal_im_detect (line 10)
[im_blob, im_scales] = get_image_blob(conf, im);

Error in script_faster_rcnn_demo (line 55)

[boxes, scores] = proposal_im_detect(proposal_detection_model.conf_proposal, rpn_net, im);

maybe the imsize function do not support gpuArray? Thanks again.

Hi @ShaoqingRen, can you tell me how many classes have you trained in 2 days on faster_rcnn?

Faster RCNN is really a fantastic work!

Recently, I want to use Faster-RCNN to help the pose estimation. The dataset of pose estimation contains about 80,000 images after data augmentation. In the stage one's do_proprosal_test() function which tries to get the new roidb, the program will require 80G+ CPU memory. I can not afford such a large memory. And it seems that if I want to use Faster RCNN in a large dataset (1 million +), it will require too much memory. So I wonder that is there any good way to save CPU memory?

Thanks a lot.

HI all,

Does Faster RCNN support multi-gpu training?
Is there anyone know how to do this?
Thanks.

I tried to run "experiments/script_faster_rcnn_demo.m" as specified in the instruction, and my matlab stopped at "output_blobs = caffe_net.forward(net_inputs);" in " proposal_im_detect.m"
I get the following error

fast_rcnn startup done
GPU 1: free memory 12538343424
GPU 2: free memory 12552810496
Use GPU 2
错误使用 caffe_
glog check error, please check log and clear mex

出错 caffe.Blob/set_data (line 27)
caffe_('blob_set_data', self.hBlob_self, data);

出错 caffe.Net/forward (line 140)
self.blobs(self.inputs{n}).set_data(input_data{n});

出错 proposal_im_detect (line 23)
output_blobs = caffe_net.forward(net_inputs);

出错 script_faster_rcnn_demo (line 54)
[boxes, scores] = proposal_im_detect(proposal_detection_model.conf_proposal, rpn_net, im);

When I run the faster_rcnn_build.m, I encountered the problem like this:
//////////////////////////////////////////////////////////////////////////
......
nms_gpu_mex.cu
mex -outdir bin nms_gpu_mex.o -L"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v7.5\lib\x64" -lcudart
libcpmt.lib(xthrow.obj) : error LNK2038: 检测到“_MSC_VER”的不匹配项: 值“1700”不匹配值“1800”(nms_gpu_mex.o 中)
libcpmt.lib(syserror.obj) : error LNK2038: 检测到“_MSC_VER”的不匹配项: 值“1700”不匹配值“1800”(nms_gpu_mex.o 中)
正在创建库 C:\Users\WENTAO1\AppData\Local\Temp\mex_qh9ZhZ\templib.x 和对象 C:\Users\WENTAO1\AppData\Local\Temp\mex_qh9ZhZ\templib.exp
nms_gpu_mex.o : error LNK2019: 无法解析的外部符号 fminf，该符号在函数 "float __cdecl fmin(float,float)" (?fmin@@yammm@Z) 中被引用
nms_gpu_mex.o : error LNK2019: 无法解析的外部符号 fmaxf，该符号在函数 "float __cdecl fmax(float,float)" (?fmax@@yammm@Z) 中被引用
......
D:\PROGRAM FILES\MATLAB\R2013A\BIN\MEX.PL: Error: Link of 'bin\nms_gpu_mex.mexw64' failed.
////////////////////////////////////////////////////////////////////////////
My platform is Matlab2013a + CUDA7.5 + VS2013compiler, how can I solve this problem?

I want to use the default detection model, except with proposal_detection_model.is_share_feature = 0, i.e. use the RPN for detection but without sharing the blobs. If I try to make that change, then it gives the error:

F1205 22:56:34.525357 40342 inner_product_layer.cpp:61] Check failed: K_ == new_K (25088 vs. 147) Input size incompatible with inner product parameters.

I'm not entirely sure why this is happening. Does the model require that the feature blobs come from the RPN, or is there a way around this?

Hello, I have a Tesla C2070 so it is 2.0 computation capability, I have lowered the number of batches and also the max number of rois. I first tried with batch with a single image and for 3000 max rois and got only 28% mAP. I just wanted to try a fast training to see if the whole process works without running into other problems. This time I train with base_lr set to 0.00001 stepsize and max_iter 7 times more than what it is in the original solver (7 because I lowered the batch size from 128 to 1 and thought I should use the exponential decrease as a multiplier) I set max number of rois to 3500 during rcnn_test this time to see if it will be okay with 2.0 (it is) but now I get a mAP of 0.0073 in total, for each object definitely around 0.00xx can't figure out why, shouldn't I have a better accuracy than my initial attempt???

Hi,
I am trying to install new caffe module provided with faster-rcnn for having my own python interface to run faster-rcnn(instead of matlab). I have protobuf package version 2.4 installed. Could anyone help me with the error given below. I am able to build fast-rcnn without any problems. Thanks in advance.

I am getting following error while importing caffe module:
Traceback (most recent call last):
File "", line 1, in
File "/home/classic/faster_rcnn/faster_rcnn-master/external/caffe/caffe-faster-R-CNN/python/caffe/init.py", line 1, in
from .pycaffe import Net, SGDSolver
File "/home/classic/faster_rcnn/faster_rcnn-master/external/caffe/caffe-faster-R-CNN/python/caffe/pycaffe.py", line 13, in
from ._caffe import Net, SGDSolver
ImportError: /home/classic/faster_rcnn/faster_rcnn-master/external/caffe/caffe-faster-R-CNN/python/caffe/../../build/lib/libcaffe.so: undefined symbol: _ZN6google8protobuf2io16CodedInputStream24default_recursion_limit_E

Thank you

I tried to run "experiments/script_faster_rcnn_demo.m" as specified in the instruction, and my matlab crashed with the following crash report

I ran the code in Ubuntu 14.04. I hope someone can shed some light on what is going on.

          abort() detected at Sat Oct 17 10:04:52 2015

Configuration:
Crash Decoding : Disabled
Current Visual : 0x20 (class 4, depth 24)
Default Encoding : UTF-8
GNU C Library : 2.19 stable
MATLAB Architecture: glnxa64
MATLAB Root : /data/mat14a
MATLAB Version : 8.3.0.532 (R2014a)
Operating System : Linux 3.13.0-65-generic #105-Ubuntu SMP Mon Sep 21 18:50:58 UTC 2015 x86_64
Processor ID : x86 Family 6 Model 63 Stepping 2, GenuineIntel
Window System : The X.Org Foundation (11701000), display localhost:11.0

Fault Count: 1

Abnormal termination:
abort()

Register State (from fault):
RAX = 0000000000000000 RBX = 00007f4c0c79b620
RCX = ffffffffffffffff RDX = 0000000000000006
RSP = 00007f4c4bf89458 RBP = 00007f4c4bf89590
RSI = 00000000000030aa RDI = 0000000000003085

R8 = 000000000000ff08 R9 = ffffffffffff1150
R10 = 0000000000000008 R11 = 0000000000000202
R12 = 0000000000000001 R13 = 00007f4c4bf897d0
R14 = 00007f4c4bf8a040 R15 = 0000000000000001

RIP = 00007f4c5e110cc9 EFL = 0000000000000202

CS = 0033 FS = 0000 GS = 0000

Stack Trace (from fault):
[ 0] 0x00007f4c5e110cc9 /lib/x86_64-linux-gnu/libc.so.6+00224457 gsignal+00000057
[ 1] 0x00007f4c5e1140d8 /lib/x86_64-linux-gnu/libc.so.6+00237784 abort+00000328
[ 2] 0x00007f4c0c575d81 /usr/lib/x86_64-linux-gnu/libglog.so.0+00068993 _ZN6google22InstallFailureFunctionEPFvvE+00000000
[ 3] 0x00007f4c0c575daa /usr/lib/x86_64-linux-gnu/libglog.so.0+00069034 ZN6google10LogMessage10SendToSinkEv+00000000
[ 4] 0x00007f4c0c575ce4 /usr/lib/x86_64-linux-gnu/libglog.so.0+00068836 ZN6google10LogMessage9SendToLogEv+00001224
[ 5] 0x00007f4c0c5756e6 /usr/lib/x86_64-linux-gnu/libglog.so.0+00067302 ZN6google10LogMessage5FlushEv+00000414
[ 6] 0x00007f4c0c578687 /usr/lib/x86_64-linux-gnu/libglog.so.0+00079495 ZN6google15LogMessageFatalD1Ev+00000025
[ 7] 0x00007f4c1657421e /home/umteht/faster_rcnn/experiments/external/caffe/matlab/+caffe/private/caffe.mexa64+01430046
[ 8] 0x00007f4c165bf823 /home/umteht/faster_rcnn/experiments/external/caffe/matlab/+caffe/private/caffe.mexa64+01738787
[ 9] 0x00007f4c16454702 /home/umteht/faster_rcnn/experiments/external/caffe/matlab/+caffe/private/caffe.mexa64+00251650
[ 10] 0x00007f4c164551c7 /home/umteht/faster_rcnn/experiments/external/caffe/matlab/+caffe/private/caffe.mexa64+00254407 mexFunction+00000154
[ 11] 0x00007f4c5603b72a /data/mat14a/bin/glnxa64/libmex.so+00120618 mexRunMexFile+00000090
[ 12] 0x00007f4c56037a94 /data/mat14a/bin/glnxa64/libmex.so+00105108
[ 13] 0x00007f4c56038fb4 /data/mat14a/bin/glnxa64/libmex.so+00110516
[ 14] 0x00007f4c55432ad9 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00670425 ZN8Mfh_file11dispatch_fhEiPP11mxArray_tagiS2+00000697
[ 15] 0x00007f4c5450920e /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02601486
[ 16] 0x00007f4c544c41d0 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02318800
[ 17] 0x00007f4c544c61ea /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02327018
[ 18] 0x00007f4c544c9167 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02339175
[ 19] 0x00007f4c544c726f /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02331247
[ 20] 0x00007f4c544c7ec4 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02334404
[ 21] 0x00007f4c5452530b /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02716427
[ 22] 0x00007f4c55432ad9 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00670425 ZN8Mfh_file11dispatch_fhEiPP11mxArray_tagiS2+00000697
[ 23] 0x00007f4c5506a6e8 /data/mat14a/bin/glnxa64/libmwmcos.so+01566440
[ 24] 0x00007f4c55014482 /data/mat14a/bin/glnxa64/libmwmcos.so+01213570
[ 25] 0x00007f4c55016465 /data/mat14a/bin/glnxa64/libmwmcos.so+01221733
[ 26] 0x00007f4c55018e50 /data/mat14a/bin/glnxa64/libmwmcos.so+01232464
[ 27] 0x00007f4c5501673d /data/mat14a/bin/glnxa64/libmwmcos.so+01222461
[ 28] 0x00007f4c5506d126 /data/mat14a/bin/glnxa64/libmwmcos.so+01577254
[ 29] 0x00007f4c550d955b /data/mat14a/bin/glnxa64/libmwmcos.so+02020699
[ 30] 0x00007f4c553e1874 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00338036 ZN13Mfh_MATLAB_fn11dispatch_fhEiPP11mxArray_tagiS2+00000244
[ 31] 0x00007f4c550d9031 /data/mat14a/bin/glnxa64/libmwmcos.so+02019377
[ 32] 0x00007f4c5450920e /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02601486
[ 33] 0x00007f4c544c41d0 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02318800
[ 34] 0x00007f4c544c61ea /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02327018
[ 35] 0x00007f4c544c9167 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02339175
[ 36] 0x00007f4c544c726f /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02331247
[ 37] 0x00007f4c544c7ec4 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02334404
[ 38] 0x00007f4c5452530b /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02716427
[ 39] 0x00007f4c55432ad9 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00670425 ZN8Mfh_file11dispatch_fhEiPP11mxArray_tagiS2+00000697
[ 40] 0x00007f4c5506a6e8 /data/mat14a/bin/glnxa64/libmwmcos.so+01566440
[ 41] 0x00007f4c5506a995 /data/mat14a/bin/glnxa64/libmwmcos.so+01567125
[ 42] 0x00007f4c55014236 /data/mat14a/bin/glnxa64/libmwmcos.so+01212982
[ 43] 0x00007f4c5501498c /data/mat14a/bin/glnxa64/libmwmcos.so+01214860
[ 44] 0x00007f4c5501638e /data/mat14a/bin/glnxa64/libmwmcos.so+01221518
[ 45] 0x00007f4c55018e50 /data/mat14a/bin/glnxa64/libmwmcos.so+01232464
[ 46] 0x00007f4c55016a9c /data/mat14a/bin/glnxa64/libmwmcos.so+01223324
[ 47] 0x00007f4c55016be9 /data/mat14a/bin/glnxa64/libmwmcos.so+01223657
[ 48] 0x00007f4c55016dcf /data/mat14a/bin/glnxa64/libmwmcos.so+01224143
[ 49] 0x00007f4c55016ff1 /data/mat14a/bin/glnxa64/libmwmcos.so+01224689
[ 50] 0x00007f4c55072717 /data/mat14a/bin/glnxa64/libmwmcos.so+01599255
[ 51] 0x00007f4c550d91f8 /data/mat14a/bin/glnxa64/libmwmcos.so+02019832
[ 52] 0x00007f4c553e1874 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00338036 ZN13Mfh_MATLAB_fn11dispatch_fhEiPP11mxArray_tagiS2+00000244
[ 53] 0x00007f4c550d9031 /data/mat14a/bin/glnxa64/libmwmcos.so+02019377
[ 54] 0x00007f4c5450920e /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02601486
[ 55] 0x00007f4c544c41d0 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02318800
[ 56] 0x00007f4c544c61ea /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02327018
[ 57] 0x00007f4c544c9167 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02339175
[ 58] 0x00007f4c544c726f /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02331247
[ 59] 0x00007f4c544c7ec4 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02334404
[ 60] 0x00007f4c5452530b /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02716427
[ 61] 0x00007f4c55432ad9 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00670425 ZN8Mfh_file11dispatch_fhEiPP11mxArray_tagiS2+00000697
[ 62] 0x00007f4c5450920e /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02601486
[ 63] 0x00007f4c544c41d0 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02318800
[ 64] 0x00007f4c544c62b0 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02327216
[ 65] 0x00007f4c544c9167 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02339175
[ 66] 0x00007f4c544c726f /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02331247
[ 67] 0x00007f4c544c8245 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02335301
[ 68] 0x00007f4c544b9a4f /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02275919
[ 69] 0x00007f4c544becb9 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02297017
[ 70] 0x00007f4c544bc979 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02287993
[ 71] 0x00007f4c544bced6 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02289366
[ 72] 0x00007f4c544b7f08 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02268936
[ 73] 0x00007f4c544b829a /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02269850
[ 74] 0x00007f4c553f4de3 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00417251
[ 75] 0x00007f4c553e1874 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00338036 ZN13Mfh_MATLAB_fn11dispatch_fhEiPP11mxArray_tagiS2+00000244
[ 76] 0x00007f4c5450920e /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02601486
[ 77] 0x00007f4c544c41d0 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02318800
[ 78] 0x00007f4c544c61ea /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02327018
[ 79] 0x00007f4c544c9167 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02339175
[ 80] 0x00007f4c544c726f /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02331247
[ 81] 0x00007f4c544c7ec4 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02334404
[ 82] 0x00007f4c5452530b /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02716427
[ 83] 0x00007f4c55432ad9 /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00670425 ZN8Mfh_file11dispatch_fhEiPP11mxArray_tagiS2+00000697
[ 84] 0x00007f4c5450920e /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02601486
[ 85] 0x00007f4c544c41d0 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02318800
[ 86] 0x00007f4c544c61ea /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02327018
[ 87] 0x00007f4c544c9167 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02339175
[ 88] 0x00007f4c544c726f /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02331247
[ 89] 0x00007f4c544c7ec4 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02334404
[ 90] 0x00007f4c5452530b /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02716427
[ 91] 0x00007f4c55432c5f /data/mat14a/bin/glnxa64/libmwm_dispatcher.so+00670815 ZN8Mfh_file11dispatch_fhEiPP11mxArray_tagiS2+00001087
[ 92] 0x00007f4c544f8135 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02531637
[ 93] 0x00007f4c544bf0d9 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02298073
[ 94] 0x00007f4c544bbdc7 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02284999
[ 95] 0x00007f4c544bc193 /data/mat14a/bin/glnxa64/libmwm_interpreter.so+02285971
[ 96] 0x00007f4c56265afc /data/mat14a/bin/glnxa64/libmwbridge.so+00142076
[ 97] 0x00007f4c56266791 /data/mat14a/bin/glnxa64/libmwbridge.so+00145297 _Z8mnParserv+00000721
[ 98] 0x00007f4c5f51c92f /data/mat14a/bin/glnxa64/libmwmcr.so+00489775 _ZN11mcrInstance30mnParser_on_interpreter_threadEv+00000031
[ 99] 0x00007f4c5f4fdb6d /data/mat14a/bin/glnxa64/libmwmcr.so+00363373
[100] 0x00007f4c5f4fdbe9 /data/mat14a/bin/glnxa64/libmwmcr.so+00363497
[101] 0x00007f4c53bf1d46 /data/mat14a/bin/glnxa64/libmwuix.so+00343366
[102] 0x00007f4c53bd4382 /data/mat14a/bin/glnxa64/libmwuix.so+00222082
[103] 0x00007f4c5fc7250f /data/mat14a/bin/glnxa64/libmwservices.so+02323727
[104] 0x00007f4c5fc7267c /data/mat14a/bin/glnxa64/libmwservices.so+02324092
[105] 0x00007f4c5fc6e57f /data/mat14a/bin/glnxa64/libmwservices.so+02307455
[106] 0x00007f4c5fc739b5 /data/mat14a/bin/glnxa64/libmwservices.so+02329013
[107] 0x00007f4c5fc73de7 /data/mat14a/bin/glnxa64/libmwservices.so+02330087
[108] 0x00007f4c5fc744c0 /data/mat14a/bin/glnxa64/libmwservices.so+02331840 _Z25svWS_ProcessPendingEventsiib+00000080
[109] 0x00007f4c5f4fe098 /data/mat14a/bin/glnxa64/libmwmcr.so+00364696
[110] 0x00007f4c5f4fe3bf /data/mat14a/bin/glnxa64/libmwmcr.so+00365503
[111] 0x00007f4c5f4f928f /data/mat14a/bin/glnxa64/libmwmcr.so+00344719
[112] 0x00007f4c5e4a7182 /lib/x86_64-linux-gnu/libpthread.so.0+00033154
[113] 0x00007f4c5e1d447d /lib/x86_64-linux-gnu/libc.so.6+01025149 clone+00000109

This error was detected while a MEX-file was running. If the MEX-file
is not an official MathWorks function, please examine its source code
for errors. Please consult the External Interfaces Guide for information
on debugging MEX-files.

If this problem is reproducible, please submit a Service Request via:
http://www.mathworks.com/support/contact_us/

A technical support engineer might contact you with further information.

Thank you for your help.

As solutions of issues told,

i uncomment 31 line of script_faster_rcnn_demo.m,

replace 7.5 to 6.5 of Cuda driver

and finally i could stop MATLAB crash

but when run script_faster_rcnn_demo,

fast_rcnn startup done
GPU 1: free memory 939069440
Use GPU 1
Error using caffe_
glog check error, please check log and clear mex

Error in caffe.Net/forward_prefilled (line 109)
caffe_('net_forward', self.hNet_self);

Error in caffe.Net/forward (line 142)
self.forward_prefilled();

Error in proposal_im_detect (line 23)
output_blobs = caffe_net.forward(net_inputs);

Error in script_faster_rcnn_demo (line 54)
[boxes, scores] =
proposal_im_detect(proposal_detection_model.conf_proposal, rpn_net, im);

this message and caffe_log.2680 file.

what should i do to run??

tell me if any necessary further information needed

(Windows 7 , VS 2010 , 6.5 Cuda driver, MATLAB 2013 and 2015)

thank you for reading

I meet the all the requirement for the system, and when I tried to run the script_faster_rcnn_demo.
I got the following errors.

Undefined function 'nms_gpu_mex' for input arguments of type 'single'.

Error in nms (line 37)
pick = nms_gpu_mex(single(boxes)', double(overlap));

Error in script_faster_rcnn_demo>boxes_filter (line 148)
aboxes = aboxes(nms(aboxes, nms_overlap_thres, use_gpu), :);

Error in script_faster_rcnn_demo (line 55)
aboxes = boxes_filter([boxes, scores], opts.per_nms_topN, opts.nms_overlap_thres, opts.after_nms_topN, opts.use_gpu);

Can anyone tell me how can I solve this problem?
By the way functions/nms folder is added on the matlab path.

matlab print info:
Preparing training data...Starting parallel pool (parpool) using the 'local' profile ... connected to 4 workers.
Done.
Preparing validation data...Done.

centos terminal printf info:
I1212 13:48:10.404944 3263 net.cpp:247] Network initialization done.
I1212 13:48:10.404949 3263 net.cpp:248] Memory required for data: 21358056
F1212 14:20:41.143805 3263 smooth_L1_loss_layer.cpp:49] Not Implemented Yet
*** Check failure stack trace: ***

this is why?
please help me!

When I run the script in cuda 7.0, I meet a error.

fast_rcnn startup done
GPU 1: free memory 7093714944
Use GPU 1
Invalid MEX-file 'D:\faster_rcnn-master\external\caffe\matlab\caffe_faster_rcnn\caffe_.mexw64': 找不到指定的模块。

Error in caffe.set_device (line 9)
caffe_('set_device', device_id);

Error in active_caffe_mex (line 26)
caffe.set_device(gpu_id-1);

Error in script_faster_rcnn_demo (line 10)
active_caffe_mex(opts.gpu_id, opts.caffe_version);

the final net model showed at the bottom，is trained by VOC 2007 trainval + 2012 trainval or VOC 2007 trainval&test + 2012 trainval（sorry to bother you by such a silly question）

Hello, I have set up faster rcnn and am able to run the demo. At the training though, at step: stage one fast rcnn I get this error "roi_pooling_layer.cu:97] Check failed: error == cudaSuccess (9 vs. 0) invalid configuration argument" I have a Tesla C2070 GPU and am using Cuda 7 which seems to be working fine. I tried to change the batch size in fast_rcnn_config and the base_lr but I still get the error. Any idea how I can fix this? Is this related to input_dim which says it is to be changed on-the-fly to num ROIs? What should I do to work around? Thanks.