caculate the accuracy of bounding box (prediction result) with groundtruth

There are many methods, such as IOU

enter image description here

It is useful for detecting objecting, but is not quite suitable for my case.

Because I just want to know the overlap lenght, so I change the caculation function.

see.

%% test the accuracy of the detection
function [overlapRatio]= CalBoundingboxAcuuracy(Mbbox,bboxes)
%Mbbox--ground truth bounding box
%bboxes--detected bounding box
%method compare the detected width /width of the groundtruth
%wd--detected groundtruth in ground truth dataset
%wt-ground truth total width
%% designed by Jun Zhang May 10, 2019 in CU ICAR

if ~isempty(Mbbox)
    wt=sum(Mbbox(:,3));
    wd=0;
    if ~isempty(bboxes)
     for i=1:size(Mbbox,1)
        %ground truth dataset
       % Mbbox(i,:);
     for j=1:size(bboxes) %the bouding box must be merged 
            %detected bbox
            %rectint(Mbbox(4,:),bboxes(j,:)); %get the area of overlap area
            %% method 1
          %  overlapRatio = bboxOverlapRatio(Mbbox(i,:),bboxes(j,:),'Min');
           % wd=overlapRatio*Mbbox(i,3)+wd;
            % if only use the overlapRatio, the result is the overlap
            % ratio,we can't caculate how much the ground truth is included
            %% method 2
           % so we use  rectint to get the overlap area
           % wd=wd+rectint(Mbbox(i,:),bboxes(j,:))/Mbbox(i,4);
           % there is still problem, because the result of rectint is their
           % overlap area,so the we can't the the overlength correctly,
           % because the boxes are not always the same.
           %% we just caculate the overlap length 
           % we judge whether they are overlaped together, then caculate
           % the length
            if bboxOverlapRatio(Mbbox(i,:),bboxes(j,:),'Min')>0
                wd=wd+min(Mbbox(i,1)+Mbbox(i,3),bboxes(j,1)+bboxes(j,3))-max(Mbbox(i,1),bboxes(j,1));
                
            end
            
          
        end
     end
     
    end
    overlapRatio= wd/wt;% include no detection ,zero
else
     
              
         overlapRatio=1;
    
        
end

评论
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包

打赏作者

做一个码农都是奢望

你的鼓励将是我创作的最大动力

¥1 ¥2 ¥4 ¥6 ¥10 ¥20
扫码支付:¥1
获取中
扫码支付

您的余额不足,请更换扫码支付或充值

打赏作者

实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值