JegZheng / CACR-SSL

Pytorch implementation of Contrastive Attraction and Contrastive Repulsion for Representation Learning
3 stars 0 forks source link

How to train cifar10 using MOCO or other self-supervised learning? #1

Open saifhassan opened 1 year ago

saifhassan commented 1 year ago

Hey @JegZheng

I implemented this repo but could not find any code to train any self-supervised method along with ResNet18 backbone on cifar10 dataset?

Thanks

JegZheng commented 1 year ago

Hi, we did not use ResNet18 or use MoCo to train models on cifar10 in our paper. We provide our code with AlexNet encoder to help reproduce results in our small-scale experiments. However, you are welcome to do it with this repo by replacing the AlexNet encoder architecture with ResNet18.

saifhassan commented 1 year ago

Hi, we did not use ResNet18 or use MoCo to train models on cifar10 in our paper. We provide our code with AlexNet encoder to help reproduce results in our small-scale experiments. However, you are welcome to do it with this repo by replacing the AlexNet encoder architecture with ResNet18.

yeah that's fine, i have trained with ResNet and other backboens, however I couldn't find any code which uses self supervised learning method such as MoCo on CIFAR10 dataset. Can you guide me in few steps, how would I train AlexNet or any backbone with MoCo or any other SSL method?

JegZheng commented 1 year ago

I think our small-scale experiments code is what you need. Please check that.

saifhassan commented 1 year ago

I think our small-scale experiments code is what you need. Please check that.

Let me explain my query. Actually I want to apply self-supervised learning method on my dataset. I have gone through your small-experiments code but it doesn't use any self-supervised learning method (such as, MoCo, SimCLR or others). I want to use SSL along with CARD model.

My Questions:

  1. Does your small experiments section use any SSL method on CIFAR10/CIFAR100 dataset?
  2. If yes, then can you point it out the file please?