dadrt's Introduction

DaDRT

Distractor-aware Deep Regression for Visual Tracking

Dependence:

* MatConvNet     ：https://github.com/vlfeat/matconvnet 
* autonn         ：https://github.com/vlfeat/autonn 
* mcnExtraLayers ：https://github.com/albanie/mcnExtraLayers 
* VGG16 model    ：http://www.vlfeat.org/matconvnet/pretrained/

dadrt's People

Contributors

Stargazers

Watchers

dadrt's Issues

lower precision plot and auc than the paper

Hello, When I run the code, the precision plot is only 93.93% in OTB-2013 which is lower than your result in your paper.
And in your code, you use Normalize(), however in mcnLayers, the output is an array, not a cell,

    function [derInputs, derParams] = backward(obj, inputs, params, derOutputs)
      ders = vl_nnscalenorm(inputs{1}, params{1}, derOutputs{1}) ;
      [derInputs, derParams] = deal(ders(1), ders(2)) ; 
    end

which can not satisfy matconvnet.

      % compute derivatives of inputs and paramerters
      [derInputs, derParams] = obj.backward ...
        (inputs, {net.params(par).value}, derOutputs) ;
      if ~iscell(derInputs) || numel(derInputs) ~= numel(in)
        error('Invalid derivatives returned by layer "%s".', layer.name);
      end

so the only modify is that:

    function [derInputs, derParams] = backward(obj, inputs, params, derOutputs)
      %ders = vl_nnscalenorm(inputs{1}, params{1}, derOutputs{1}) ;
      [derI, derP] = vl_nnscalenorm(inputs{1}, params{1}, derOutputs{1}) ;
     derInputs{1}  = derI;
     derParams{1} = derP;
      %[derInputs, derParams] = deal(ders(1), ders(2)) ; 
    end