菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS)

一、MCG-ICT-CAS Object Detection

1、获得成绩

菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS)_第1张图片

2、overview

菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS)_第2张图片

3、主要的创新点

(1)在建立tubelet时,提出了基于detection的检测方式

a.Adjacent Checking–By optical-flow for precise tubelet

   If IoU (Red, Green) > a given threshold: same tubelet 

    else: a new tubelet

b.Detection Box Sequentialization

c.Coherent reclassification–Use majority voting to get coherent categories

   throughout a given tubelet

菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS)_第3张图片

但是发现效果不够好,会出现目标消失和轨迹离散的情况,于是想到将基于检测的tubelet与基于track的tubelet相结合


tracking based tubelet:(作者发现会出现位置漂移)

a.Anchor Frame Selection–Select the frame with highest detection score object(其他方法大部分都是用的这个方法)

b.Anchor Target Selection–Exploit the adjacent information with optical flow [5]to determine     the reliable anchor targets

菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS)_第4张图片

c.Anchor Target Selection: remove the unreliable去掉一些不可相信的检测框(通过bbox         match)

d.Multi-target tracking with detection recall


(2)tubelet fusion:

有两种方式:

Union Fusion——Merge overlapping tubelets(有重叠部分)

菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS)_第5张图片
Union Fusion

Concatenation Fusion——Merge successive tubelets(连接的)

菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS)_第6张图片

4、其他结构配置

still image detection:

菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS)_第7张图片

你可能感兴趣的:(菜鸟实习日记~day21(ILSVRC2016之MCG-ICT-CAS))