This page walks through the steps required to run MYOW on the CIFAR10 dataset.
Training is parallalised using DistributedDataParallel
. The pool of candidate views during mining is shared across
all instances.
To start training run:
CUDA_VISIBLE_DEVICES=0,1 python3 scripts/cifar-train.py \
--lr 2.0 \
--mm 0.98 \
--weight_decay 5e-5 \
--optimizer sgd \
--lr_warmup_epochs 30 \
--batch_size 256 \
--port 12354 \
--logdir myow-cifar \
--ckptpath myow-cifar
Evaluation can be done simultaneously or after training on a separate GPU instance. The eval script will automatically
run evaluation each time a new checkpoint is saved to ckptpath
. It is also possible to start evaluation only after
a certain number of epoch using the resume_eval
argument.
CUDA_VISIBLE_DEVICES=2 python3 scripts/cifar-eval.py \
--lr 0.04 \
--resume_eval 0 \
--logdir runs-cifar \
--ckptpath myow-cifar
tensorboard --logdir=runs-cifar