GOTURN Tracker results in errors

[bassamarshad]

System information (version)

• OpenCV => 3.2.0 & opencv_contrib 3.2
• Operating System / Platform => Windows 10 64 Bit
• Compiler => Visual Studio 2015

Detailed description

I have been trying to run the GOTURN tracker OpenCV implementation. I was firstly not able to find the goturn.prototxt and goturn.caffemodel (pretrained versions) on the opencv_extra github site, but was able to pull them from other github repos.

The program fails on execution with the following error, as soon as you select a ROI :

OpenCV Error: Assertion failed (input.dims() == 4 && (input.type() == CV_32F || input.type() == CV_64F)) in cv::dnn::ConvolutionLayerImpl::allocate, file E:\opencv3.2.0_sources\opencv_contrib-master\modules\dnn\src\layers\convolution_layer.cpp, line 90
OpenCV Error: Assertion failed (The following error occured while making allocate() for layer "conv11": input.dims() == 4 && (input.type() == CV_32F || input.type() == CV_64F)) in cv::dnn::ConvolutionLayerImpl::allocate, file E:\opencv3.2.0_sources\opencv_contrib-master\modules\dnn\src\layers\convolution_layer.cpp, line 90

I tried fixing this by updating the following lines of code in gtrtacker.cpp as follows :

dnn::Blob targetBlob = dnn::Blob(targetPatch);
dnn::Blob searchBlob = dnn::Blob(searchPatch);

TO

dnn::Blob targetBlob = dnn::Blob::fromImages(targetPatch);
dnn::Blob searchBlob = dnn::Blob::fromImages(searchPatch);

NOW,

I get a different error altogether that seems to have something to do with OpenCL, i get this error in the allocate() function in dnn.cpp ::

OpenCV Error: Assertion failed (dims <= 3) in cv::ocl::OpenCLAllocator::checkContinuous, file E:\opencv3.2.0_sources\3.2.0\opencv-3.2.0\modules\core\src\ocl.cpp, line 4899
OpenCV Error: Assertion failed (The following error occured while making forward() for layer "concat1": dims <= 3) in cv::ocl::OpenCLAllocator::checkContinuous, file E:\opencv3.2.0_sources\3.2.0\opencv-3.2.0\modules\core\src\ocl.cpp, line 4899
0x00007FFD18AC7788 (0x000000AAD070BBF8 0x000000AAD070BC00 0x000000AAD070BC08 0x00007FFD18D38CC1), RaiseException() + 0x68 bytes(s)
0x00007FFCF5F74582 (0x00007FFCDBED0000 0x000000AAD071BD50 0x0000021C8281D384 0x00007FFD18D30000), _CxxThrowException() + 0xC2 bytes(s)
0x00007FFCDC53A45A (0x0000000000000029 0x0000000000000000 0x0000000000000000 0x0000000000000000), cv::error() + 0x15A bytes(s), e:\opencv3.2.0_sources\3.2.0\opencv-3.2.0\modules\core\src\system.cpp, line 661 + 0x1E byte(s)
0x00007FFCDC53A5A0 (0x000000AAD071E3D0 0x0000000000000000 0x000000AAD071CB90 0x0000021C81BB9E64), cv::error() + 0x140 bytes(s), e:\opencv3.2.0_sources\3.2.0\opencv-3.2.0\modules\core\src\system.cpp, line 666 + 0x109 byte(s)
0x00007FFCDD71E173 (0x00007FFCDD71E0F0 0x000000AAD072ECD0 0x000000AAD072ECD0 0x0000000000000000), cv::dnn::Net::Impl::forwardLayer'::1'::catch$0() + 0x83 bytes(s), e:\opencv3.2.0_sources\3.2.0\opencv_contrib-3.2.0\modules\dnn\src\dnn.cpp, line 423 + 0x75 byte(s)
0x00007FFCF5F7C720 (0x00007FFCDD71E0F0 0x000000AAD071D1A8 0x0000000000000100 0x0000000000000000), __C_specific_handler() + 0x230 bytes(s)
0x00007FFCF5F72AE2 (0x0000000000000000 0x0000000000000000 0x000000AAD071E250 0x0000000000000000), __FrameUnwindFilter() + 0x432 bytes(s)
0x00007FFD1B94A193 (0x0000021CEA99E8B0 0x0000021CEAADBE08 0x0000021CE832FFE0 0x0000021C8222F7F0), RtlCaptureContext() + 0x3C3 bytes(s)
0x00007FFCDD601006 (0x0000021CEAADB720 0x0000021CEAADB748 0x0000000000000001 0x0000021CE8330008), cv::dnn::Net::Impl::forwardLayer() + 0x156 bytes(s), e:\opencv3.2.0_sources\3.2.0\opencv_contrib-3.2.0\modules\dnn\src\dnn.cpp, line 419 + 0x1B byte(s)
0x00007FFCDD600E4F (0x0000000000000000 0x000000AAD072ED90 0x0000021CE828CE70 0x0000021CE832FFE0), cv::dnn::Net::Impl::forwardAll() + 0x8F bytes(s), e:\opencv3.2.0_sources\3.2.0\opencv_contrib-3.2.0\modules\dnn\src\dnn.cpp, line 435
0x00007FFCDD600D7E (0x0000000000000046 0x000000AAD072EFA8 0x000000AAD072EEB0 0x0000000000000001), cv::dnn::Net::forward() + 0x9E bytes(s), e:\opencv3.2.0_sources\3.2.0\opencv_contrib-3.2.0\modules\dnn\src\dnn.cpp, line 503
0x00007FFCDD85CB5E (0x000000C3CD82EBAD 0x0000000000000001 0x00007FF679A68700 0x000000C3CD82EBAD), cv::gtr::TrackerGOTURNImpl::updateImpl() + 0xA5E bytes(s), e:\opencv3.2.0_sources\3.2.0\opencv_contrib-3.2.0\modules\tracking\src\gtrtracker.cpp, line 170 + 0x31 byte(s)
0x00007FF679A61FFB (0x00007FFD18E169F8 0x0000000000000000 0x00007FFD18E16A08 0x0000000000000000), main() + 0x5FB bytes(s), e:\tracking_goturn\tracking_goturn\source.cpp, line 168 + 0x1B byte(s)
0x00007FF679A632E8 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), __scrt_common_main_seh() + 0x124 bytes(s), f:\dd\vctools\crt\vcstartup\src\startup\exe_common.inl, line 264 + 0x22 byte(s)
0x00007FFD1AC98364 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), BaseThreadInitThunk() + 0x14 bytes(s)
0x00007FFD1B9070D1 (0x0000000000000000 0x0000000000000000 0x0000000000000000 0x0000000000000000), RtlUserThreadStart() + 0x21 bytes(s)

Not sure if there is a problem in my environment -_ i tried reverting builds to 3.1.0 , with no success. Or maybe I dont have the correct *.prototxt and *.caffemodel.

Am using the model and prototxt file given here :
https://github.com/opencv/opencv_extra/tree/c4219d5eb3105ed8e634278fad312a1a8d2c182d/testdata/tracking

Let me know what you guys think !

Steps to reproduce

[lenlen]

/cc @Auron-X

[Auron-X]

@bassamarshad Thank you for detailed description of the bug in GOTURN. I will try to fix it ASAP

[dev681999]

Well apparently using #include <opencv2/core/ocl.hpp> this and using heaocl::setUseOpenCL(false); like 1st line in main() solves the first problem mentioned.
Don't know about second one

[serhan-gul]

I'm getting exactly the same error on OpenCV 3.2.0 & opencv_contrib 3.2.0 (Ubuntu 16.04) after the selection of a bounding box:

OpenCV Error: Assertion failed (input.dims() == 4 && (input.type() == CV_32F || input.type() == CV_64F)) in allocate, file /home/guel/opencv_contrib/modules/dnn/src/layers/convolution_layer.cpp, line 87
OpenCV Error: Assertion failed (The following error occured while making allocate() for layer "conv11": input.dims() == 4 && (input.type() == CV_32F || input.type() == CV_64F)) in allocate, file /home/guel/opencv_contrib/modules/dnn/src/layers/convolution_layer.cpp, line 87
terminate called after throwing an instance of 'cv::Exception'
  what():  /home/guel/opencv_contrib/modules/dnn/src/layers/convolution_layer.cpp:87: error: (-215) The following error occured while making allocate() for layer "conv11": input.dims() == 4 && (input.type() == CV_32F || input.type() == CV_64F) in function allocate

Could it be related to the .prototxt and .caffemodel files from opencv_extra repository? I used the same files as basamarshad.

[Auron-X]

Hello. I am still working on this bug, apparently there is some issue with dnn/tracking modules used in combo. For now you can try GOTURN outside of the OpenCV tracking API, using an example on my Github: https://github.com/Auron-X/GOTURN-Example
It also work on GPU ~100 FPS (on my GTX 1080) as the original paper stated

[allensu0314]

Thanks to @LorenaGdL for the PR about this bug, I didn't meet this issue with latest code.
But I met another error below:
OpenCV Error: Requested object was not found (Requested blob ".data1" not found) in cv::dnn::Net::setBlob, file C:\opencv3_2\opencv_contrib-master\modules\dnn\src\dnn.cpp, line 516
It seems that there's something wrong in gtrTracker.cpp line 167:

    dnn::Blob targetBlob = dnn::Blob::fromImages(targetPatch);
    dnn::Blob searchBlob = dnn::Blob::fromImages(searchPatch);

   (line 167) net.setBlob(".data1", targetBlob);
    net.setBlob(".data2", searchBlob);

I just download *.ptototxt and *.caffemodel, and put them to my VS program folder. Then add "path to test video" in Command Arguments. BTW, my main.cpp is below:

#include <opencv2/core/utility.hpp>
#include <opencv2/tracking.hpp>
#include <opencv2/videoio.hpp>
#include <opencv2/highgui.hpp>
#include <iostream>
#include <cstring>
using namespace std;
using namespace cv;
int main(int argc, char** argv){
	// show help
	if (argc<2){
		cout <<
			" Usage: tracker <video_name>\n"
			" examples:\n"
			" example_tracking_kcf Bolt/img/%04d.jpg\n"
			" example_tracking_kcf faceocc2.webm\n"
			<< endl;
		return 0;
	}
	// declares all required variables
	Rect2d roi;
	Mat frame;
	// create a tracker object
	Ptr<Tracker> tracker = Tracker::create("GOTURN");
	// set input video
	std::string video = argv[1];
	VideoCapture cap(video);
	// get bounding box
	cap >> frame;
	roi = selectROI("tracker", frame);
	//quit if ROI was not selected
	if (roi.width == 0 || roi.height == 0)
		return 0;
	// initialize the tracker
	tracker->init(frame, roi);
	// perform the tracking process
	printf("Start the tracking process, press ESC to quit.\n");
	for (;;){
		// get frame from the video
		cap >> frame;
		// stop the program if no more images
		if (frame.rows == 0 || frame.cols == 0)
			break;
		// update the tracking result
		tracker->update(frame, roi);
		// draw the tracked object
		rectangle(frame, roi, Scalar(255, 0, 0), 2, 1);
		// show image with the tracked object
		imshow("tracker", frame);
		//quit on ESC button
		if (waitKey(1) == 27)break;
	}
	return 0;
}

I don't know how to fix it, I will be appreciate if anyone can help me.

[xiaochus]

@ChaselS314 I have the same problem with you when i use GOTURN on latest code.

It seems that getPinByAlias() return a invalid variable pin when parser the '.data1' by splitPin() in dnn.cpp line 296&266:

    LayerPin getPinByAlias(const String &pinAlias, bool isOutPin = true)
    {
        LayerPin pin;
        String layerName, outName;
        splitPin(pinAlias, layerName, outName);

        pin.lid = (layerName.empty()) ? 0 : getLayerId(layerName);

        if (pin.lid >= 0)
            pin.oid = resolvePinOutputName(getLayerData(pin.lid), outName, isOutPin);

        return pin;
    }

Therefore， setBlob() raise an error about ‘.data1’.

void Net::setBlob(String outputName, const Blob &blob)
{
    LayerPin pin = impl->getPinByAlias(outputName);
    if (!pin.valid())
        CV_Error(Error::StsObjectNotFound, "Requested blob \"" + outputName + "\" not found");

    LayerData &ld = impl->layers[pin.lid];
    ld.outputBlobs.resize( std::max(pin.oid+1, (int)ld.requiredOutputs.size()) );
    ld.outputBlobs[pin.oid] = blob;
}

But I am not familliar with C++ so i do not know how to solve it.

[puneetdabulya]

Still getting this error on OpenCV 3.2 and Ubuntu 14.04, CUDA 8.0 Titan X GPU

[hacunamatada]

@ChaselS314, @xiaochus OpenCV Error: Requested object was not found (Requested blob ".data1" not found) in cv::dnn::Net::setBlob, file C:\opencv3_2\opencv_contrib-master\modules\dnn\src\dnn.cpp, line 516

The above error seems to be caused by the mismatching of the goturn.prototxt file.
I received the file from the following site.
Https://github.com/opencv/opencv_extra/tree/c4219d5eb3105ed8e634278fad312a1a8d2c182d/testdata/tracking
The following command merge four files into one.
cat goturn.caffemodel.zip.001 goturn.caffemodel.zip.002 goturn.caffemodel.zip.003 goturn.caffemodel.zip.004 > goturn.caffemodel
However, an error still occurs.
The error message is as follows.
OpenCV Error: Assertion failed ((bias && l-> blobs.size () == 2) || (! Bias && l-> b Lobs.size () == 1)) in cv :: dnn :: initConvDeconvLayerFromCaffe, file D: \ Program \ ope Ncv-dnn \ opencv_contrib-master \ opencv_contrib-master \ modules \ dnn \ src \ caffe \ layer_ Loaders.cpp, line 30

in my case : l-> blobs.size () is 0.

[isgursoy]

[libprotobuf ERROR /opt/opencv/320deb/modules/dnn/3rdparty/protobuf/sources/protobuf-3.1.0/src/google/protobuf/text_format.cc:298] Error parsing text-format caffe.NetParameter: 7:1: Expected identifier, got: <
OpenCV Error: Unspecified error (FAILED: ReadProtoFromTextFile(param_file, param). Failed to parse NetParameter file: goturn.prototxt) in ReadNetParamsFromTextFileOrDie, file /opt/opencv/opencv_contrib-3.2.0/modules/dnn/src/caffe/caffe_io.cpp, line 1101
/opt/opencv/opencv_contrib-3.2.0/modules/dnn/src/caffe/caffe_io.cpp:1101: error: (-2) FAILED: ReadProtoFromTextFile(param_file, param). Failed to parse NetParameter file: goturn.prototxt in function ReadNetParamsFromTextFileOrDie

OpenCV Error: Unspecified error (GOTURN network loading error...) in cv::gtr::InitImpl, file gtrTracker.cpp, line 117
terminate called after throwing an instance of 'cv::Exception'
what(): gtrTracker.cpp:117: error: (-2) GOTURN network loading error... in function cv::gtr::InitImpl

[ahrnbom]

I get the following exotic errors when trying to run GOTURN via OpenCV:

OpenCV Error: Assertion failed ((bias && l->blobs.size() == 2) || (!bias && l->blobs.size() == 1)) in initConvDeconvLayerFromCaffe, file /opencv/opencv_contrib-3.2.0/modules/dnn/src/caffe/layer_loaders.cpp, line 27 OpenCV Error: Assertion failed (The following error occured while making allocate() for layer "conv11": (bias && l->blobs.size() == 2) || (!bias && l->blobs.size() == 1)) in initConvDeconvLayerFromCaffe, file /opencv/opencv_contrib-3.2.0/modules/dnn/src/caffe/layer_loaders.cpp, line 27 Traceback (most recent call last): File "pipeline.py", line 32, in <module> ok, bbox = tracker.update(im) cv2.error: /opencv/opencv_contrib-3.2.0/modules/dnn/src/caffe/layer_loaders.cpp:27: error: (-215) The following error occured while making allocate() for layer "conv11": (bias && l->blobs.size() == 2) || (!bias && l->blobs.size() == 1) in function initConvDeconvLayerFromCaffe

I am also a bit confused by the comment here https://github.com/opencv/opencv_contrib/blob/master/modules/tracking/samples/goturnTracker.cpp where it says

//Demo of GOTURN tracker
//In order to use GOTURN tracker, GOTURN architecture goturn.prototxt and goturn.caffemodel are > required to exist in root folder.
//There are 2 ways to get caffemodel:
//1 - Train you own GOTURN model using https://github.com/Auron-X/GOTURN_Training_Toolkit
//2 - Download pretrained caffemodel from https://github.com/opencv/opencv_extra

As the OpenCV Extra repo doesn't seem to contain any Goturn caffemodel, pretrained or otherwise. I downloaded it from http://cs.stanford.edu/people/davheld/public/GOTURN/trained_model/tracker.caffemodel instead. Is that is what is intended, or is it some other .caffemodel file one is supposed to use?

[LaurentBerger]

I think something is weird in those lines. In my goturn.prototxt there is no .data1 and .data2 field only data1 and data2

if I delete point before data then exception given by @lChaselS314 doest not occur :

OpenCV Error: Requested object was not found (Requested blob ".data1" not found) in cv::dnn::Net::setBlob, file C:\opencv3_2\opencv_contrib-master\modules\dnn\src\dnn.cpp, line 516

my goturn.prototxt file is

name: "GOTURN"

input: "data1"
input_dim: 1
input_dim: 3
input_dim: 227
input_dim: 227

input: "data2"
input_dim: 1
input_dim: 3
input_dim: 227
input_dim: 227

layer {
  name: "conv11"
  type: "Convolution"
  bottom: "data1"
  top: "conv11"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 4
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu11"
  type: "ReLU"
  bottom: "conv11"
  top: "conv11"
}
layer {
  name: "pool11"
  type: "Pooling"
  bottom: "conv11"
  top: "pool11"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm11"
  type: "LRN"
  bottom: "pool11"
  top: "norm11"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv12"
  type: "Convolution"
  bottom: "norm11"
  top: "conv12"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu12"
  type: "ReLU"
  bottom: "conv12"
  top: "conv12"
}
layer {
  name: "pool12"
  type: "Pooling"
  bottom: "conv12"
  top: "pool12"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm12"
  type: "LRN"
  bottom: "pool12"
  top: "norm12"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv13"
  type: "Convolution"
  bottom: "norm12"
  top: "conv13"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu13"
  type: "ReLU"
  bottom: "conv13"
  top: "conv13"
}
layer {
  name: "conv14"
  type: "Convolution"
  bottom: "conv13"
  top: "conv14"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu14"
  type: "ReLU"
  bottom: "conv14"
  top: "conv14"
}
layer {
  name: "conv15"
  type: "Convolution"
  bottom: "conv14"
  top: "conv15"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu15"
  type: "ReLU"
  bottom: "conv15"
  top: "conv15"
}
layer {
  name: "pool15"
  type: "Pooling"
  bottom: "conv15"
  top: "pool15"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}


layer {
  name: "conv21"
  type: "Convolution"
  bottom: "data2"
  top: "conv21"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 96
    kernel_size: 11
    stride: 4
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu21"
  type: "ReLU"
  bottom: "conv21"
  top: "conv21"
}
layer {
  name: "pool21"
  type: "Pooling"
  bottom: "conv21"
  top: "pool21"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm21"
  type: "LRN"
  bottom: "pool21"
  top: "norm21"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv22"
  type: "Convolution"
  bottom: "norm21"
  top: "conv22"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 2
    kernel_size: 5
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu22"
  type: "ReLU"
  bottom: "conv22"
  top: "conv22"
}
layer {
  name: "pool22"
  type: "Pooling"
  bottom: "conv22"
  top: "pool22"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}
layer {
  name: "norm22"
  type: "LRN"
  bottom: "pool22"
  top: "norm22"
  lrn_param {
    local_size: 5
    alpha: 0.0001
    beta: 0.75
  }
}
layer {
  name: "conv23"
  type: "Convolution"
  bottom: "norm22"
  top: "conv23"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "relu23"
  type: "ReLU"
  bottom: "conv23"
  top: "conv23"
}
layer {
  name: "conv24"
  type: "Convolution"
  bottom: "conv23"
  top: "conv24"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 384
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu24"
  type: "ReLU"
  bottom: "conv24"
  top: "conv24"
}
layer {
  name: "conv25"
  type: "Convolution"
  bottom: "conv24"
  top: "conv25"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  convolution_param {
    num_output: 256
    pad: 1
    kernel_size: 3
    group: 2
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu25"
  type: "ReLU"
  bottom: "conv25"
  top: "conv25"
}
layer {
  name: "pool25"
  type: "Pooling"
  bottom: "conv25"
  top: "pool25"
  pooling_param {
    pool: MAX
    kernel_size: 3
    stride: 2
  }
}

layer {
	name: "concat1"
	type: "Concat"
	bottom: "pool15"
	bottom: "pool25"
	top: "poolConcat"
}

layer {
  name: "fc6"
  type: "InnerProduct"
  bottom: "poolConcat"
  top: "fc6"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 4096
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu6"
  type: "ReLU"
  bottom: "fc6"
  top: "fc6"
}
layer {
  name: "drop6"
  type: "Dropout"
  bottom: "fc6"
  top: "fc6"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc7"
  type: "InnerProduct"
  bottom: "fc6"
  top: "fc7"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 4096
    weight_filler {
      type: "gaussian"
      std: 0.005
    }
    bias_filler {
      type: "constant"
      value: 1
    }
  }
}
layer {
  name: "relu7"
  type: "ReLU"
  bottom: "fc7"
  top: "fc7"
}
layer {
  name: "drop7"
  type: "Dropout"
  bottom: "fc7"
  top: "fc7"
  dropout_param {
    dropout_ratio: 0.5
  }
}
layer {
  name: "fc8"
  type: "InnerProduct"
  bottom: "fc7"
  top: "fc8"
  param {
    lr_mult: 1
    decay_mult: 1
  }
  param {
    lr_mult: 2
    decay_mult: 0
  }
  inner_product_param {
    num_output: 4
    weight_filler {
      type: "gaussian"
      std: 0.01
    }
    bias_filler {
      type: "constant"
      value: 0
    }
  }
}
layer {
  name: "scale"
  bottom: "fc8"
  top: "out"
  type: "Power"
  power_param {
    power: 1
    scale: 10
    shift: 0
  }
}