Comments (23)
@Linsongrong 您好,GCN-V的目标是让每个vertex只输出一个confidence,用来表示其属于特别类别的置信度,故而nclass设置为1
from learn-to-cluster.
@yl-1993 @Linsongrong 你好,检测了下GCN-V输入的是feature_dim=256,但我们提取的特征集都是512,而看你们都是提取512维特征,这个是不是相互矛盾呢?向各位求解
from learn-to-cluster.
from learn-to-cluster.
@yl-1993 @Linsongrong ,再次请教你们
目前由于自身训练的特征提取模型是512的,所有想训练基于512特征的GCN_V和GCN_E模型,使用的也是face_emore 5.8M的训练集,目前参考cfg_train_gcnv_ms1m来布置训练,但得到的损失一直为 losss: nan , 微调lr=0.01以后损失在很快降低到0.009以下,训练20000次后,用训练集进行测试分值在0.9以上,不同阈值分类数量差很多,这是运用10W小数据,试跑的情况,现在开始尝试5.8M数据训练
@yl-1993 请问作者和大家,我该如何配置参数训练,已得到好的结果呢
from learn-to-cluster.
from learn-to-cluster.
from learn-to-cluster.
@Linsongrong
嗯,感谢你的回复! 其他参数没动,改了feature_dim=512,lr=0.05,会有损失值,若默认为 lr=0.1则出现损失 loss = nan
尝试训练5.8M的数据集,以提高模型泛化能力,但完全走不动,请问你是怎么制定数据集大小的呢?,可以根据什么准确计算得到数据集限制大小呢
from learn-to-cluster.
from learn-to-cluster.
from learn-to-cluster.
@Linsongrong 嗯,正在尝试。你目前对于512GCNV网络训练取得的成绩怎样呢,引用做参考
from learn-to-cluster.
from learn-to-cluster.
@Linsongrong @yl-1993
对于256特征我有以下探讨意见
基于256特征维度提取到的人脸特征,用作人脸识别鲁棒性没有512特征维度强,用作聚类进行衔接项目,降低维度来进行聚类,有两种方法:1、是训练输出256特征网络,但会降低模型总体能力,从大数据上考量,即使很小值也会产生非常大的数量差距;2、从512特征上稀疏到256上来进行聚类,这个没尝试过,不知道你是否有试过
个人分析对于512特征聚类相比于256聚类的劣势是速度上会有所降低,运算资源占用相对较多,但优势在取得分数上会相对容易得到高分,且最终结果也相对会高一些,可能不多,这受限于模型的复杂度
from learn-to-cluster.
@XHQC 你好, 你是对的,你对GCNV重新进行512维训练出结果了吗?效果如何?相比256维,时间花费大概增长了多少?
from learn-to-cluster.
@Linsongrong 256的模型我没有,故没有测试,512的模型我测试的结果是15~20分钟 50W数据,基于GCN-V,由于配置原因不稳定,
from learn-to-cluster.
from learn-to-cluster.
@Linsongrong 精度应该是有优势的比256模型,我的测试结果显示。 我想做5000W数据的聚类,你有没有进行过分批聚类呢,通过怎样的方式进行?
from learn-to-cluster.
from learn-to-cluster.
@Linsongrong 对于分批聚类再聚类你是怎样选择再聚类数据呢,是通过选择最大置信度顶点再聚类嘛,或者是?
from learn-to-cluster.
from learn-to-cluster.
@Linsongrong 类中心的计算方式,你是通过类均值中心值或者最近的点来选取的吧? 而置信度顶点是模型推导出的点,可以试试是否可以获取更好的类中心,最优类中心的定义我认为是正脸清晰的照片,通过计算方法得来的类中心可能会造成偏离最优类中心,这个问题貌似不好避免,由类内数据分布不均造成
from learn-to-cluster.
@yl-1993 @Linsongrong @XHQC 感谢大家的真知灼见。我在使用自有数据集微调训练gcn-v的过程中发现train loss忽上忽下(可能跟我选用adam优化器有关),基本最后都会过拟合(train loss较低,但test loss超级高),而且test loss最小的模型,评估出来的指标很低啊,感觉loss失去了指导模型训练的作用。那么该怎么选择模型呢?难道每次迭代的模型都保存评估一遍?我试了下,发现在测试集指标FP能达到87%,但是其loss不是最低也不是最高,看不出规律。
from learn-to-cluster.
@XHQC 你好,方便给个邮箱地址吗,我训练上有些问题想向你请教一下。
from learn-to-cluster.
@Linsongrong 256的模型我没有,故没有测试,512的模型我测试的结果是15~20分钟 50W数据,基于GCN-V,由于配置原因不稳定,
有512的模型么,可否共享一下,我用来测试一下聚类效果,我的特征是512维
from learn-to-cluster.
Related Issues (20)
- output error HOT 1
- Dataset issue HOT 2
- Issue while training LGCN HOT 1
- 您好,想请教一下训练GCN-D时报错run out of input HOT 1
- 没有修改任何参数跑作者提供的pretrained model for ms1m Fscore 只有0.32
- Clustering 512D features using pretrained model HOT 3
- building symmetric adjacency matrix in utils HOT 1
- why using output of relu for rebuilding knn graph HOT 1
- is not a checkpoint filemodels/pretrained_gcn_e_ms1m.pth HOT 3
- about extracting features HOT 3
- About how to make a .meta file HOT 1
- Can we manually set the number of clusters during the test HOT 1
- 大佬,请教下train_cluster_seg_iop_vox.sh 的test阶段报错的问题 HOT 2
- 两个face recognition模型,推荐使用哪个,hfsoftmax还要用到python 2 ? HOT 1
- About pretrained_model to get the the extracted feature HOT 2
- GCN-D MMdistributedDataparallel train HOT 2
- 关于性能 HOT 1
- How to apply lgcn for clustering on custom embeddings of collected face images? HOT 1
- On custom dataset preparation
- On the numpy version which may cause exception HOT 3
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
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.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from learn-to-cluster.