search

DonkeyShot21 / cassle

Official repository for the paper "Self-Supervised Models are Continual Learners" (CVPR 2022)
MIT License
117 stars 18 forks source link

Need of checkpoints : BT and VicReg #5

Closed KameniAlexNea closed 1 year ago

KameniAlexNea commented 2 years ago

Hi all,

Is there any link where I can access to the checkpoint of model trained using Barlowtwins and VicREG ?

I would like to evaluate this approach using different models and need the trained last checkpoint of these models.

Thanks.

vturrisi commented 2 years ago

Hey. I'm a bit busy this week, but I'll try to share the checkpoints as soon as possible.

KameniAlexNea commented 1 year ago

Hi, can you please share the configuration used to obtain results in assets\results file ?

I have tried to change some confs without getting it. Updating learning rate as mentioned in the paper.

Are you using online evaluation when training ? At the end of CSSL, what percentage of data are you using to train your linear model when trained on CIFAR-100 ?

DonkeyShot21 commented 1 year ago

Hi, you can find the script in bash_files/continual/{dataset}/{method}.sh. They should contain the correct arguments.

On CIFAR100 we report the online linear eval accuracy since offline eval does not improve. Except for table 9 (semi-supervised), we use 100% of the data to train the linear classifier.

KameniAlexNea commented 1 year ago

Thank you for your feedback.

II apologize for the omission in the information provided. I think that this argument could be added in the files on CIFAR-100.