nicewsyly/eco

c++ visual studio implement of ECO: Efficient Convolution Operators for Tracking

eco's Issues

怎么不用caffe,只调用HC特征？

这个caffe版我觉得多此一举，无gpu每秒8帧的话也太慢了，eco hc版在哪里有么？

关于fDSST中，“scale filter”问题咨询

您好，很高兴看到优秀的C++代码实现。我在看作者提供的原始版本的代码时，一个问题一直困扰着我。
1）在fDSST中，作者用如下代码生成ys, 在调试中发现，ys两边值最大，中间值最小，而在检测过程中，作者也是找响应最大的位置作为当前的scale。
2）在DSST中，将中间的值设置为最大，两边的值较小是，与fDSST相冲突。
如果能提供任何意见，将非常感激。
ys = [1 0.642167199144790 0.170056201828270 0.0185709050136830 0.000836315026159023 1.55311269123150e-05 1.18941188747878e-07 3.75627879293239e-10 4.89192285304428e-13 4.89192285304428e-13 3.75627879293239e-10 1.18941188747878e-07 1.55311269123150e-05 0.000836315026159023 0.0185709050136830 0.170056201828270 0.642167199144790]

scale_exp = (-floor((nScales-1)/2):ceil((nScales-1)/2)) * nScalesInterp/nScales;
    scale_exp_shift = circshift(scale_exp, [0 -floor((nScales-1)/2)]);
    
    interp_scale_exp = -floor((nScalesInterp-1)/2):ceil((nScalesInterp-1)/2);
    interp_scale_exp_shift = circshift(interp_scale_exp, [0 -floor((nScalesInterp-1)/2)]);
    
    scaleSizeFactors = scale_step .^ scale_exp;
    interpScaleFactors = scale_step .^ interp_scale_exp_shift;
    
    ys = exp(-0.5 * (scale_exp_shift.^2) /scale_sigma^2);

ubuntu16.04 环境下编译通过，但运行出错

首先感谢nicewsyly，开源ECO c++版本给我们学习。
我在ubuntu16.04编译通过，但是运行时出错了，不管是否使用cnn特征都一样的错误，我使用的是960x540的视频，错误提示如下如下：
init obj[486 216 46 101]
img_support_sz is : [252 x 252]
eco track...0
process_frame 1
process_frame 2
process_frame 3
process_frame 4
process_frame 5
process_frame 50
process_frame 51
process_frame 52
terminate called after throwing an instance of 'cv::Exception'
what(): OpenCV(3.4.2) /home/mrsy/project/machine-learning-lib/opencv-3.4.2/modules/core/src/matrix.cpp:465: error: (-215:Assertion failed) 0 <= roi.x && 0 <= roi.wid
th && roi.x + roi.width <= m.cols && 0 <= roi.y && 0 <= roi.height && roi.y + roi.height <= m.rows in function 'Mat'

打印信息的代码如下：

抗干扰效果怎么样？

当跟踪目标被部分遮挡，或者傍边有相似目标时，会跟错吗

The Rect Don't move？

When I run, why doesn't it feel good, and the box doesn't move, is it my fault?

Freeze in both CPU and GPU mode when deep features in use

eco.log

Nice work though not runnable for me

I am trying my best to make your codes runnable the whole afternoon,but unfortunatly I failed,as although in you describination the caffe-windows is a CPU version,some of codes suffer from a CUDA-related error that I have to uncommon them to let the compiler pass, besides the caffe-windows I used did not contains some of your additional dependencies such as caffe.lib.lib, let alone cuda.lib.
Would you plz make the USAGE more detailed so that anyone interested in your work would ultimate and contribute it.
And thanks for your sharing.

useDeepFeature为1的话，运行到net.reset(new Net<float>(proto, TEST))出就会退出。

useDeepFeature=0能正常运行，这时候的特征是hog还是hog+cn？

GPU版本？

您好，请问有GPU版本吗？

What's the speed of this C++ version?

Great work! What's the speed (FPS) of this C++ version?

new commit uploaded can run on gpu version of caffe ?

Crash in non deep features mode

Variable k1 is uninitialized.
https://github.com/martin-danelljan/ECO/search?q=k1&unscoped_q=k1

追踪不正确

我在Ubuntu中测试, 使用HOG版本.
output_sz=0, k1=0, frameID=0, frames_since_last_train=0都初始化为0.
在测试视频中, 发现追踪有比较大问题, 不知道是哪里需要修正?

執行一陣子會關閉!?

請問這樣的狀況是!?

How about the effect when the targets are multi_persons?

我这里跑k1没初始化

ECO.cpp
大概110-120行之间：
for (size_t i = 0; i != feature_sz.size(); ++i)
{
size_t size = feature_sz[i].width + (feature_sz[i].width + 1) % 2 ;
filter_sz.push_back(cv::Size(size, size));
k1 = size > output_sz ? i : k1;
//wangsen
output_sz = std::max(size,output_sz);
//output_sz = size > output_sz ? size : output_sz;
}
output_sz好像也是随机的值，然后一直大于size，k1一直没得初始化，

When does the algorithm stop tracking the target?

thanks for your work, i dont find the judgment that will stop tracking the target? thanks

既然已经用了caffe_cpu版本，为什么不用caffe_gpu版本？

或者说是否我自己比较容易实现替换成gpu版本的caffe？

compile in linux

time efficiency?

Thanks for sharing your work~
It seems not as fast as the Matlab version ? ( ~ 6 fps vs. 30 fps on my PC )
I complied CPU version and turn useDeepFeature=false, imshow is also turned off.
Is there something I missed? Thanks~

fhog accelerate

I analyzed the program.and "Fhog" it takes a long time.

Is there a way to speed up?

About ECO execution

After I compile program successfully, I find some files are lost.
I downloaded VGG_CNN_M_2048.caffemodel and VGG_mean.binaryproto, but I can't find the file which named mean.yml.
How can I do?

The initialization of output_sz

Looks like there is no initialization of the variable output_sz in ECO class?

I tried to run in HOG version and looks like it is not correctly initialized so there is a segmentation fault.