GithubHelp home page GithubHelp logo

Training Problem about trackit HOT 7 CLOSED

researchmm avatar researchmm commented on June 3, 2024
Training Problem

from trackit.

Comments (7)

JudasDie avatar JudasDie commented on June 3, 2024

Hi, thanks for your work, I have some problem in training.
I find that in TracKit/experiments/train/Ocean.yaml this model train for 50 epoch, but in the paper training for 20 epoch, and the learing rate end_lr also seems for alexnet training.
Another problem is that I only have one 2080ti gpu, will it influence the results a lot? Or should I adjust some parameters.
Thanks.

Hi. 1) The full training is performed for 50 epochs, as detailed in Sec.5.1 in the paper. The Sec.5.3 provides the abolition for training 20 epochs. When conducting your idea on this framework, you could first train 20 epoch for fast evaluation. After getting a good result, then do the full train. 2) One GPU may decrease the performance since the batch size is much smaller than 256. You could try to increase the batch size.

from trackit.

paozhuanyinyuba avatar paozhuanyinyuba commented on June 3, 2024

hello, 在数据集一样,用代码中提供的设置文件,用八卡训练的效果。并不能达到论文中的效果,可能是什么原因呢? 相差还比较大

from trackit.

JudasDie avatar JudasDie commented on June 3, 2024

hello, 在数据集一样,用代码中提供的设置文件,用八卡训练的效果。并不能达到论文中的效果,可能是什么原因呢? 相差还比较大

Hi, 请问所有配置都一样么?可以展示下你的log么?另外你epoch test后结果是多少(VOT19)?以及调参后结果是多少?

from trackit.

paozhuanyinyuba avatar paozhuanyinyuba commented on June 3, 2024

hello, 在数据集一样,用代码中提供的设置文件,用八卡训练的效果。并不能达到论文中的效果,可能是什么原因呢? 相差还比较大

Hi, 请问所有配置都一样么?可以展示下你的log么?另外你epoch test后结果是多少(VOT19)?以及调参后结果是多少?

配置就用的 experiments/train/Ocean.yaml, 在VOT2018上的结果是 EAO 0.357 (这是没有 tune 参数的结果)

from trackit.

JudasDie avatar JudasDie commented on June 3, 2024

hello, 在数据集一样,用代码中提供的设置文件,用八卡训练的效果。并不能达到论文中的效果,可能是什么原因呢? 相差还比较大

Hi, 请问所有配置都一样么?可以展示下你的log么?另外你epoch test后结果是多少(VOT19)?以及调参后结果是多少?

配置就用的 experiments/train/Ocean.yaml, 在VOT2018上的结果是 EAO 0.357

1)如果方便的话麻烦传给我一份train/epoch test/tune的log,我帮忙check下,如果github不能传的话请发我邮箱[email protected] 2) 请问可以同时反馈一下VOT19的结果么?

from trackit.

paozhuanyinyuba avatar paozhuanyinyuba commented on June 3, 2024

hello, 在数据集一样,用代码中提供的设置文件,用八卡训练的效果。并不能达到论文中的效果,可能是什么原因呢? 相差还比较大

Hi, 请问所有配置都一样么?可以展示下你的log么?另外你epoch test后结果是多少(VOT19)?以及调参后结果是多少?

配置就用的 experiments/train/Ocean.yaml, 在VOT2018上的结果是 EAO 0.357

1)如果方便的话麻烦传给我一份train/epoch test/tune的log,我帮忙check下,如果github不能传的话请发我邮箱[email protected] 2) 请问可以同时反馈一下VOT19的结果么?

行 我找时间整理下, 我觉着问题是不是出现在 bn上, 我粗略看了下好像没有用synbn; EAO 0.357是没有 tune para的结果

from trackit.

JudasDie avatar JudasDie commented on June 3, 2024

hello, 在数据集一样,用代码中提供的设置文件,用八卡训练的效果。并不能达到论文中的效果,可能是什么原因呢? 相差还比较大

Hi, 请问所有配置都一样么?可以展示下你的log么?另外你epoch test后结果是多少(VOT19)?以及调参后结果是多少?

配置就用的 experiments/train/Ocean.yaml, 在VOT2018上的结果是 EAO 0.357

1)如果方便的话麻烦传给我一份train/epoch test/tune的log,我帮忙check下,如果github不能传的话请发我邮箱[email protected] 2) 请问可以同时反馈一下VOT19的结果么?

行 我找时间整理下, 我觉着问题是不是出现在 bn上, 我粗略看了下好像没有用synbn; EAO 0.357是没有 tune para的结果

应该不是,我之前复现波动没有这么大,也麻烦你顺便测下19,因为18比较容易过拟合,19和20相对稳定,用来测试可能更容易反映tracker性能(个人经验仅供参考)。也麻烦你整理完邮箱发我下,我也找时间帮你check。

from trackit.

Related Issues (20)

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.