Coder Social home page Coder Social logo

Comments (5)

HikariTJU avatar HikariTJU commented on August 28, 2024
  1. 在retina和fcos这几个上面只用了 general distribution,没有使用dfl和qfl
  2. retina的bbox_coder实际上没用到

from ld.

Logicino avatar Logicino commented on August 28, 2024
  1. 在retina和fcos这几个上面只用了 general distribution,没有使用dfl和qfl
  2. retina的bbox_coder实际上没用到

感谢回复~!
可以理解为采用generalized distribution是loss_bbox=dict(type='GIoULoss')这个决定的吗?
另外,关于蒸馏实验,我之前采用您的方法做实验,确实对于同样使用ResNet+FPN的情况下,用ResNet101指导ResNet18有很大提高
但是这几天尝试了使用MobileNet作为student,用ResNet-101的结果作为teacher会有一些精度下降,不知道是不是蒸馏对网络结构有要求,即网络结构近似的效果会更好?
(但是我理解的蒸馏是将teacher网络的分类结果,这里是推演出的边框概率位置,作为指导学生的,所以感觉可能是没关系的?是不是应该往调整蒸馏学习的超参数方向试试呢?谢谢~!

from ld.

Zzh-tju avatar Zzh-tju commented on August 28, 2024

准确来说有三个超参值得考虑,一个是边界被离散化成概率分布的离散程度n (小区间的个数),gfocal是16,对于rotated RetinaNet我们发现8更好,因此n可能会有一定的影响,不过差别不会太大。

第二个是温度\tau,这影响着概率分布的软化程度。一般而言会在5~15这个范围比较好。

第三个是loss weight,这个可以调一调。

我建议是按照这个顺序逐一去确定。

from ld.

HikariTJU avatar HikariTJU commented on August 28, 2024

没做过mobilenet的实验,你可以用resnet18或者34做teacher试试

from ld.

Logicino avatar Logicino commented on August 28, 2024

准确来说有三个超参值得考虑,一个是边界被离散化成概率分布的离散程度n (小区间的个数),gfocal是16,对于rotated RetinaNet我们发现8更好,因此n可能会有一定的影响,不过差别不会太大。

第二个是温度\tau,这影响着概率分布的软化程度。一般而言会在5~15这个范围比较好。

第三个是loss weight,这个可以调一调。

我建议是按照这个顺序逐一去确定。

谢谢您的回复!我这几天重新做了实验,发现teacher选择上也有一定影响
之前用res18做student的话,LD之后效果最好的是res50
mobile我试了还是res101效果最好,猜测可能因为res101下AP^S最高,补足了mobile的AP^S低的问题
我再按您的方法试一试~感谢!

from ld.

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.