- Create a conda virtual environment and activate it.
conda create --name ood2023 python=3.8.13
conda activate ood2023
conda install pytorch==1.13.1 torchvision==0.14.1 -c pytorch
- Install the dependent libraries.
pip install -r requirements.txt
- Download the data set from the official website and organize it into the following data structure.
dataset/OODCV2023
train
Images
aeroplane/
...
train/
labels.csv
phase1-test-images
phase2-test-images
- We use one Nvidia Tesla 4090 (24G) GPU for training. You need to modify the params in the
configs/beitv2.yaml
file or configs/eva02.yaml
before training, like: data
model
or pretrained weight path in pretrained_cfg
. Run different bash files for different models.
./train_eva02.sh 0 or ./train_beitv2.sh 0
- You need to modify the
checkpoint_list
in the eval_tta.py
file before testing.