Git Product home page Git Product logo

Comments (17)

Z-Xiong avatar Z-Xiong commented on July 17, 2024

特征融合或特征交互,目前常用的有:
SiamRPN++和Ocean用的Depth-Wise Cross Correlation
ATOM和DIMP用的Feature Modulation
TransT和TrDIMP用的Transformer
综合性能最好的是Transformer,但速度和部署都有问题;Correlation速度快,部署也能解决。因此nanotrack和LightTrack一样,也选择了Correlation。就是有点类似模式识别里面的滑窗卷积,为了找到最相似的某一块地方,或者说是模板匹配。
可以参考这篇知乎:https://zhuanlan.zhihu.com/p/419900331

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

好的好的,非常感谢您的指引

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

您好,又来打扰您了,请问您在跑nanotrack的时候,那位博主的test.py或者eval.py文件您能运行嘛,我总是有这个错误,调试的过程看的不是很懂哪里出错了,希望得到您的指点,感谢!
image

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

image

from lighttrack.

Z-Xiong avatar Z-Xiong commented on July 17, 2024

你单步调试下看看为什么有nan

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

您好,感谢您的建议,我昨晚调试后发现了问题,再请教您一个问题,就是nanotrack这里为什么需要对原图片进行裁剪之后再进入训练,这里的271以及511是什么意思,刚入门vot这块,麻烦您了,我在看siamfc的代码的时候好像都是直接把数据集放进去训练的。
image
image

from lighttrack.

Z-Xiong avatar Z-Xiong commented on July 17, 2024

这是数据预处理不同导致的,他们使用了不同的训练框架。271和511是对图像目标边界扩充后的大小。271应该是template size,511是search size。simfc里面训练时候也做了这样的边界扩展,只是没有把图像裁剪保存。

from lighttrack.

Z-Xiong avatar Z-Xiong commented on July 17, 2024

你可以看看nanotrack使用的原始训练框架,里面的流程很清楚。https://github.com/STVIR/pysot
另外学习单目标跟踪我不建议从nanotrack开始,建议从pysot或者pytracking这种比较成熟的训练框架入手。

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

好的好的,非常感谢,因为之前尝试去部署nanotrack的想从lighttrack作为baseline,但是苦于没有train的部分,所以直接就看了nanotrack,非常感谢您的指引

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

您好,又来打扰您了,首先感谢您的指引,我最近在看pysot,发现确实nanotrack就是按照pysot的框架写的,我在train pysort的SiamRPN++的时候,发现pysot裁剪的都是511大小的,还是上次那个问题,就是这里的271以及511,是对边界的扩展,但是为什么有两个值?这里的271以及511究竟用哪个裁剪过的大小进入训练。现在很疑惑,我看推理的过程都是将第一帧的boundingbox裁剪后,resize到127或者255,那么在训练的过程中,到底是哪个尺寸进入训练,希望得到您的解答,非常感谢
image

from lighttrack.

Z-Xiong avatar Z-Xiong commented on July 17, 2024

你先理解template size和search size。目标跟踪是有两个输入图片的。跟目标检测不一样。

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

对,train的时候是输入两个图片,我看运行par_crop.py裁剪之后,得到的是一帧的两个图片,一个template,一个serach,我就是搞不清楚的是为啥这里par_crop给的参数是511,不应该输入的是两张图片嘛(1个127,1个255)
image

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

您好,又来打扰您了,非常感谢您之前的帮助,以及您的热心,请问您有没有尝试取去复现nanotrack,我在训练的时候,准备了数据集,但是不知道VIDEOS_PER_EPOCH这个参数应该设置为多少,看了pysort官方四个数据集VID+DET+YT-BB+COCO设置的是600000,但是我在设置为600000发现不论是单独加载got还是6个数据集都放进去,训练的效果是没有改变的,eao只有0.260左右,希望能够得到您的解答,非常感谢您🙏
image

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

from lighttrack.

Z-Xiong avatar Z-Xiong commented on July 17, 2024

我没有训练过nanotrack 喔。

from lighttrack.

Coder-TieHH avatar Coder-TieHH commented on July 17, 2024

我没有训练过nanotrack 喔。

好的好的,非常感谢,我想做无人机的小目标跟踪,您有推荐的baseline吗,孩子毕不了业了/(ㄒoㄒ)/~~

from lighttrack.

Z-Xiong avatar Z-Xiong commented on July 17, 2024

我没有训练过nanotrack 喔。

好的好的,非常感谢,我想做无人机的小目标跟踪,您有推荐的baseline吗,孩子毕不了业了/(ㄒoㄒ)/~~

加微信x545690660

from lighttrack.

Related Issues (3)

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.