Improving Transferability of Representations via Augmentation-Aware Self-Supervision
Accepted to NeurIPS 2021
TL;DR: Learning augmentation-aware information by predicting the difference between two augmented samples improves the transferability of representations.
Dependencies
conda create -n AugSelf python=3.8 pytorch=1.7.1 torchvision=0.8.2 cudatoolkit=10.1 ignite -c pytorch
conda activate AugSelf
pip install scipy tensorboard kornia==0.4.1 sklearn
Checkpoints
We provide ImageNet100-pretrained models in this Dropbox link.
Pretraining
We here provide SimSiam+AugSelf pretraining scripts. For training the baseline (i.e., no AugSelf), remove --ss-crop and --ss-color options. For using other frameworks like SimCLR, use the --framework option.
STL-10
CUDA_VISIBLE_DEVICES=0 python pretrain.py \
--logdir ./logs/stl10/simsiam/aug_self \
--framework simsiam \
--dataset stl10 \
--datadir DATADIR \
--model resnet18 \
--batch-size 256 \
--max-epochs 200 \
--ss-color 1.0 --ss-crop 1.0
ImageNet100
python pretrain.py \
--logdir ./logs/imagenet100/simsiam/aug_self \
--framework simsiam \
--dataset imagenet100 \
--datadir DATADIR \
--batch-size 256 \
--max-epochs 500 \
--model resnet50 \
--base-lr 0.05 --wd 1e-4 \
--ckpt-freq 50 --eval-freq 50 \
--ss-crop 0.5 --ss-color 0.5 \
--num-workers 16 --distributed
Evaluation
Our main evaluation setups are linear evaluation on fine-grained classification datasets (Table 1) and few-shot benchmarks (Table 2).
linear evaluation
CUDA_VISIBLE_DEVICES=0 python transfer_linear_eval.py \
--pretrain-data imagenet100 \
--ckpt CKPT \
--model resnet50 \
--dataset cifar10 \
--datadir DATADIR \
--metric top1
few-shot
CUDA_VISIBLE_DEVICES=0 python transfer_few_shot.py \
--pretrain-data imagenet100 \
--ckpt CKPT \
--model resnet50 \
--dataset cub200 \
--datadir DATADIR
30 Nov 23, 2022
111 Dec 31, 2022
65 Dec 22, 2022
264 Jan 02, 2023
5 Mar 14, 2022
9 Sep 23, 2022
184 Dec 18, 2022
2 Jan 03, 2022
1 Nov 21, 2021
26 Nov 28, 2022
62 Dec 03, 2022
71 Dec 25, 2022
17 Dec 14, 2022
343 Jan 03, 2023
85 Jan 02, 2023
689 Dec 25, 2022
3 Jun 03, 2022
0 Nov 12, 2021
168 Dec 29, 2022
11 Dec 13, 2022