- Ubuntu 22.04
- Python 3.10 with pip
Install dependencies:
$ pip3 install -r requirements.txtTrain on CIFAR-10:
$ python3 main.py fit --config=config/hsvit_cifar10.yamlTest on CIFAR-10:
$ python3 main.py test --config=config/hsvit_cifar10.yaml --ckpt_path="lightning_logs/version_0/checkpoints/epoch=0-step=38035.ckpt"Train ResNet18 on CIFAR-10:
$ python3 main.py fit --config=config/sota_cifar10.yaml --model.init_args.sota_config_path="./config/huggingface/resnet-18.json"Download Hugginface ResNet50 model (pytorch_model.bin and config.json) to ./huggingface/ folder. Then, train HSViT + ResNet50 on ImageNet-1k:
$ python3 main.py fit --config=config/hsvit_het_imagenet.yamlStart a TensorBoard service based on directory of lightning_logs:
$ pip3 install tensorboard
$ tensorboard --logdir=lightning_logs/Then go to https://localhost:6006 on your web browser for viewing the training logs.
RuntimeError: received 0 items of ancdataRef: GitHub Issue
Increase the ulimit:
import resource
rlimit = resource.getrlimit(resource.RLIMIT_NOFILE)
resource.setrlimit(resource.RLIMIT_NOFILE, (2048, rlimit[1]))If you find the code useful, please use the following BibTeX entry.
@misc{xu2024hsvit,
title={HSViT: Horizontally Scalable Vision Transformer},
author={Chenhao Xu and Chang-Tsun Li and Chee Peng Lim and Douglas Creighton},
year={2024},
eprint={2404.05196},
archivePrefix={arXiv},
primaryClass={cs.CV}
}