marsggbo / MICCAI2022-EMARS

[MICCAI2022] Evolutionary Multi-objective Architecture Search Framework: Application to COVID-19 3D CT Classification

[MICCAI2022] Evolutionary-Multi-objective-Architecture-Search-Framework: Application to COVID-19 3D CT Classification

He Xin¹, Guohao Ying², Jiyong Zhang³, and Xiaowen Chu^14
1 Hong Kong Baptist University, Hong Kong, China
² University of Southern California, CA, USA
³ School of Automation, Hangzhou Dianzi University, Hang Zhou, China
⁴ The Hong Kong University of Science and Technology (Guangzhou), China

Install

pip install -r requirements.txt

search

You can refer to scripts/search_ct.sh for more run scripts.

CUDA_VISIBLE_DEVICES=0 python search.py --config_file ./configs/search.yaml logger.name MyExp

retrain

You can refer to scripts/retrain_ct.sh

there are two mode for retraininig：

• you can manually choose a promising architecture by specifying --arch_path to the path of json file, e.g., output/MyExp/version_0/epoch_66.json, and then run the following command

CUDA_VISIBLE_DEVICES=0 python retrain.py --config_file ./configs/retrain.yaml --arc_path outputs/MyExp/version_0/epoch_66.json input.size [128,128]

• the second is to finetune each selected candidate architecture for a few epochs, and then choose the best-performing one for further training. In this case, you can specify --arc_path to the log path, e.g., output/MyExp/version_0. The json files in this path will be loaded automatically:

CUDA_VISIBLE_DEVICES=0 python retrain.py --config_file ./configs/retrain.yaml --arc_path outputs/MyExp/version_0  input.size [128,128]