Multiple Object Stitching for Unsupervised Representation Learning

1. Requirements

conda create -n mos python=3.8
pip install -r requirements.txt

2. Datasets

Torchvision provides CIFAR10, CIFAR100 datasets. The root paths of data are respectively set to ./dataset/cifar10 and ./dataset/cifar100.

3. Trained Model Weights & Accuracy

Dataset	Training (#Epochs)	ViT-Tiny/2	ViT-Small/2	ViT-Base/2
CIFAR10	Pretrain (800)	link	link	link
	KNN Accuracy	93.2%	95.1%	95.1%
	Linear (100)	link	link	link
	Linear Accuracy	94.8%	96.3%	96.4%
	Finetune (100)	link	link	link
	Finetune Accuracy	97.6%	98.3%	98.3%
CIFAR100	Pretrain (800)	link	link	link
	KNN Accuracy	67.8%	73.5%	74.4%
	Linear (100)	link	link	link
	Linear Accuracy	73.5%	78.5%	79.6%
	Finetune (100)	link	link	link
	Finetune Accuracy	83.7%	86.1%	86.2%

4. Usage: Pretraining

ViT-Small with 1-node (8-GPU) training

Set hyperparameter, dataset and GPUs in config/pretrain/vit_small_pretrain.py and run the following command

python main_pretrain.py --arch vit-small

5. Usage: KNN

Set hyperparameter, dataset and GPUs in config/knn/knn.py and run the following command

python main_knn.py --arch vit-small --pretrained-weights /path/to/pretrained-weights.pth

6. Usage: Linear Classification

Set hyperparameter, dataset and GPUs in config/linear/vit_small_linear.py and run the following command:

python main_linear.py --arch vit-small --pretrained-weights /path/to/pretrained-weights.pth

7. Usage: End-to-End Fine-tuning

Set hyperparameter, dataset and GPUs in config/finetuning/vit_small_finetuning.py and run the following command

python python main_finetune.py --arch vit-small --pretrained-weights /path/to/pretrained-weights.pth