kaist-dmlab / mqnet Goto Github PK

Do you have any plan for making the requirements.txt file, in order to quickly set up the environment?

Hello author,
Thanks for the release of the code of your paper.

For ImageNet 1K, you used 50 classes as ID and other 950 classes as OOD.
If you don't mind, could you perform other options such as (ID: 100, OOD:900 or ID: 500, OOD:500)?

Thank you in advance! ^_^

How to determine the In-Out distribution ratio of each dataset?

Hi, I sincerely appreciate you sharing your code with the community.

While attempting to reproduce your results, I encountered an issue where I consistently achieved 3-5% lower accuracy across each baseline method, as shown below:

I followed the exact configuration specified in the Readme. Here are the details:

python3 main_split.py --epochs 200 --epochs-csi 1000 --epochs-mqnet 100 --datset 'CIFAR10' --n-class 10 --n-query 500 \
                      --method 'MQNet' --mqnet-mode 'LL' --ssl-save True --ood-rate 0.6

I suspect the issue might be related to the versions of the packages used. Could you kindly share the specific environment and package versions you used to achieve your published results?

Here's the environment I used, for reference:

mqnet.txt

Thank you for your assistance!

I would like to set up the environment to run MQNet. Is there a docker image for the environment?

what is the meaning of --ssl-save in running script?

Where can I find the dataset source used in the paper, especially the ImageNet that seems unable to download via torchvision?

Do you have any plans to create code using TensorFlow?

Do you have any plan to make run.sh file?

Is it available with PyTorch 2.0?

I wonder if there is a reason and effect for self validation.

is it possible to implement skyline regularation in Tensorflow version 2?

can you share visualization code for informativeness and purity score?

For the provided code,
--epochs 200 --epochs-csi 1000 --epochs-mqnet 100
how are these numbers set?

any plans for a setup.py file?