lukasruff / deep-svdd Goto Github PK
View Code? Open in Web Editor NEWRepository for the Deep One-Class Classification ICML 2018 paper
License: MIT License
Repository for the Deep One-Class Classification ICML 2018 paper
License: MIT License
I've noticed when using hard margin for SVDD, you're resetting R after training each epoch according to the outlier ratio nu.
Written in the code comments:
'set R to be the (1-nu)-th quantile of distances'
However, in
Deep-SVDD/src/opt/sgd/train.py
Line 102 in f3bebc4
Thanks.
Hello,
I would like to compare with your method in the gtsrb STOP sign database. Could you please direct me to the exact version you tested in your paper? Thanks.
Pram
Hi author,
Are you still planning to release the tensorflow version?
I'm looking forward to see it.
Thank you!
(What the title says)
Hi sir
When I run the main.py code,it has a mistake ---ValueError: 'default' must be a list when 'multiple' is true.
I think it's the version of 'Click' module, but I don't know why, looking for your reply,thanks a lot.
Hello Authors
I have followed the exact steps mentioned in the guide to setup the working environment
I am working on Ubuntu 18.10 VM on virtual box within mac
Could you please help how to resolve this error below
AssertionError: AbstractConv2d Theano optimization failed: there is no implementation available supporting the requested options. Did you exclude both "conv_dnn" and "conv_gemm" from the optimizer? If on GPU, is cuDNN available and does the GPU support it? If on CPU, do you have a BLAS library installed Theano can link against?
I want to run soft-boundary Deep SVDD,but it looks like there's only an implementation of algorithm one-class Deep SVDD. How to run soft-boundary Deep SVDD?
Thanks for the great work! May I ask if you can provide any guidelines or tutorials on how to use Deep SVDD on custom image datasets?
Any suggestion is highly appreciated!
Hi, sir. Thanks for your excellent idea!
Now I download your code and re-implement your experiment. But I get tow different results when all parameters are same in two experiment:
sh scripts/cifar10_cae.sh gpu cifar_cae 0 adam 0.0001 350 3 1 -1
sh scripts/cifar10_svdd.sh gpu cifar 0 adam 0.0001 150 0.1 1 1 0 cifar_cae/weights_final 3 1 -1
In first model, I got result:
Train objective: 0.12176
Train accuracy: 90.03%
Val objective: 0.11618
Val accuracy: 91.10%
Test objective: 0.13841
Test accuracy: 23.58%
Test AUC: 62.60%
and then, I set all parameters of the second model exactly the same as the first model, I got the result:
Train objective: 0.12050
Train accuracy: 90.02%
Val objective: 0.11537
Val accuracy: 90.80%
Test objective: 0.13548
Test accuracy: 22.60%
Test AUC: 62.23%
I got the same situation in other classes of cifar-10 dataset. In some class, the difference between the results of two identical models can even reach 2.3%.
I think that something wrong happened, but I don‘t know how to improve the result. I am looking forward to your reply!
Thanks for your great work!
There is a point that I do not understand clearly. Since that the objective function of One-Class Deep SVDD (OC-SVDD) is
Thanks a lot for your reply!
Dear author.
In MNIST, we use the original training and test splits in our experiments and only train with training set examples from the respective normal class. When I use IF in sklearn package, i find when i choose 0 as normal class, i get auroc is 0.9296.Did I get it wrong?
We look forward to your reply,thank you very much.
Hi, Sir!
Thanks for your excellent idea! I download your code and plan to recreate your experiment results. But when i set the parameters with
'sh scripts/cifar10_svdd.sh gpu cifar 0 adam 0.0001 150 0.1 1 1 0 exp 3 1 -1 '
,
I got a bad result
Train objective: 1.69638
Train accuracy: 90.02%
Val objective: 1.75854
Val accuracy: 88.90%
Test objective: 3.78288
Test accuracy: 31.02%
Test AUC: 60.71%
I think that something wrong happened, but I dont know how to improve the result. I am looking forward to your reply!
I'm sorry to bother you, but I'd like to know how you generated the boundary attack adversarial samples of the stop dataset. Or what tools did you use.I'm very confused about this and I hope to get your advice and discuss with you.
Hello. I have a question regarding the number of epochs you are training for. In the paper, you mention that DCAE is trained for "250+100 epochs" and DSVDD is trained for "150+100" epochs.
Does this mean that:
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.