Git Product home page Git Product logo

Comments (13)

ChesonHuang avatar ChesonHuang commented on June 8, 2024

给下你的valid_texts.jsonl中的2-3条数据看看,我之前遇到这样的问题是,在json里面,一个文本只对应了少部分数据,但实际一个文本应该对应所有数据。例如,在验证集中,如果你的一个文本,对应了 5 个图片,那么这5个图片都要放在image_ids里面,你可以看实现逻辑:即 ground_truth_ids里面的数据少了,所以导致recall分数低
image

from chinese-clip.

EasonTuT avatar EasonTuT commented on June 8, 2024

@ChesonHuang 我get到你的意思,但是我用的数据集中一张图片只对应一个image_ids,train、valid和test都是如此,而在fitfune过程中ACC是显示80多的,我用R@1计算出来更低只有0.几分,下面是我的训练记录和对应的json情况。
这是微调记录:
image
下面是valid的json:
image
下面是train的json:
image
我不知道什么原因导致如此ACC和R@间的分数相差如此之大

from chinese-clip.

keminze avatar keminze commented on June 8, 2024

验证集和训练集的ACC都达到了80+,但是在测试集和验证集中R@5分只有18分和9分,请问是什么问题呢?应该如何解决呢?

请问你的参数设在是怎样的

from chinese-clip.

EasonTuT avatar EasonTuT commented on June 8, 2024

@ChesonHuang @keminze
image
这是我运行脚本得出来的召回分数
是用微调后的权重的

from chinese-clip.

EasonTuT avatar EasonTuT commented on June 8, 2024

@keminze
image

from chinese-clip.

timecoderr avatar timecoderr commented on June 8, 2024

@ChesonHuang @keminze image 这是我运行脚本得出来的召回分数 是用微调后的权重的

你好,我遇到了同样的问题,请问您发现错误 解决了吗? 感谢!

from chinese-clip.

ChesonHuang avatar ChesonHuang commented on June 8, 2024

@ChesonHuang @keminze image 这是我运行脚本得出来的召回分数 是用微调后的权重的

你好,我遇到了同样的问题,请问您发现错误 解决了吗? 感谢!

建议你在 cn_clip/eval/evaluation.py 或 cn_clip/eval/evaluation_tr.py, 打印下 ground_truth_ids 和 top10_pred_ids 看看,或许它可以告诉真相

from chinese-clip.

hkxxxxx avatar hkxxxxx commented on June 8, 2024

Snipaste_2024-04-21_18-46-40
dan'qi'shi但其实我输出后也就是这样,应该没问题吧,因为他标签都是一对一的

from chinese-clip.

ChesonHuang avatar ChesonHuang commented on June 8, 2024

Snipaste_2024-04-21_18-46-40 dan'qi'shi但其实我输出后也就是这样,应该没问题吧,因为他标签都是一对一的

有问题的,你的图片id是: 140010041888, 但你模型预测的top10里面没有这个图片id,所以r@x都是0;说明你训练的模型预测不准确

update: 还有一种可能是,输入的图片必须存在valid数据集对应文本的image_ids里面,但140010041888这个图片并没有在image_ids中

from chinese-clip.

hkxxxxx avatar hkxxxxx commented on June 8, 2024

嗯,应该是模型训练的问题,查140010041888上面是有的,不然他就会报错了

from chinese-clip.

keminze avatar keminze commented on June 8, 2024

@ChesonHuang @keminze image 这是我运行脚本得出来的召回分数 是用微调后的权重的

你的训练集和验证集的文本处理方式是一样的吗

from chinese-clip.

hkxxxxx avatar hkxxxxx commented on June 8, 2024

是一样的,我一起处理的

from chinese-clip.

hkxxxxx avatar hkxxxxx commented on June 8, 2024

所以他这个Acc到底是怎么计算的?是什么的准确率?Acc和R@N之间有必然的联系吗

from chinese-clip.

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.