michalislazarou / ilpc Goto Github PK

Hi, I found that the features for WRN-28-10 need access to be downloaded, I have submitted a request on Google Drive, could you please pass it? Thank you very much! Also, when I ran the demo (python transductive_resnet12.py --dataset miniImagenet --algorithm ilpc --alpha 0.8 --K 15 --n_shots 1 --which_dataset pkl) I found that the following two imports do not execute correctly, they may be unnecessary?

1. from data_closer.datamgr import SetDataManager, TransformLoader
2. from models import res_mixup_model, wrn_mixup_model
   Thanks for your attention.

Hi,
thanks for the good work.
How to download your checkpoints here https://drive.google.com/drive/folders/1KfPzwMvVzybvp13IQW5ipHvSxBncTA-C??
I did many requests and got nothing.

Great job guys! I am reading your paper and code, however, I am having problems understanding what is going with the parameter no_samples, what is it for?

sorry, I cannot run the demo you provide. In test_arguments.py, from pkl_dataset.transform_cfg import transforms_test_options, transforms_list seems not exist in your source codes. Thanks for your attention.

Greetings,
I've been trying to reproduce the reported results in your paper. However, when I run python semi_resnet12.py --dataset miniImagenet --algorithm ilpc --alpha 0.8 --K 15 --n_shots 1, I find that the label_denoising function cost ~50 minutes on a single episode, which means >800 hours on 1000 episodes to obtain the final results.
The reason lies in that there are 1000 iterations within the label_denoising function.
I wonder if you ever face the same problems? Is there any method to boost the computation?
Thanks