search

Yunfan-Li / Contrastive-Clustering

Code for the paper "Contrastive Clustering" (AAAI 2021)
MIT License
297 stars 91 forks source link

Low Accuracy and NMI #36

Open mrsadeghi95-preteckt opened 2 years ago

mrsadeghi95-preteckt commented 2 years ago

Hi, Thank you for sharing your code. I would like to reproduce your results on CIFAR-10, I ran your original code with 4 GPUs and the results are attached below. My final ACC (NMI) after 990 epochs is about 69% (64%). Did you use any special method and/or hyperparameters for training your network which is not uploaded on Github? I would appreciate it if you could help me to reproduce your results. result_batch_256.txt

Yunfan-Li commented 2 years ago

I guess it is due to the multi-gpu implementation. How are the ICH and CCH losses computed in your multi-gpu implementation?

mrsadeghi95-preteckt commented 2 years ago

I didn't change them, I just added some commands for converting the network and data loader to multiple gpus. I can share the code with you if you like.

Yunfan-Li commented 2 years ago

In contrastive learning, samples need to be gathered from other GPUs to compute the InfoNCE loss. You may refer to our new repo https://github.com/Yunfan-Li/Twin-Contrastive-Learning for the multi-gpu implementation.