off-line linear evaluation

[allenhaozhu]

I did an evaluation on the barlow-twin model (ResNet18 on CIFAR ) provided by your checkpoint. The result is 91.56 acc@1. So is there anything wrong？ I used python3 ../../main_linear.py --dataset cifar10 --backbone resnet18 --data_dir ./ --max_epochs 100 --gpus 0 --sync_batchnorm --precision 16 --optimizer sgd --scheduler step --lr 0.1 --lr_decay_steps 60 80 --weight_decay 0 --batch_size 128 --num_workers 4 --name general-linear-eval --pretrained_feature_extractor cifar/trained_models/barlow_twins/barlow-cifar10-otu5cw89-ep\=999.ckpt --project self-supervised --wandb

[vturrisi]

Hey, The numbers that we provide for cifar are from the online evaluation. In our experiments, these are usually a little higher than what you would get with the standard offline evaluation. Probably if you train for longer or tune the parameters, you will get the same performance.

[vturrisi]

I'm closing this, if you need anything else feel free to re-open.