Skip to content

Latest commit

 

History

History
43 lines (33 loc) · 1.13 KB

get_started.md

File metadata and controls

43 lines (33 loc) · 1.13 KB
Raw

Build SIATune

Dockerfile (RECOMMENDED)

docker build . -t siatune:main -f docker/Dockerfile

Build From Source

# 1. Install PyTorch
pip install torch torchvision -f https://download.pytorch.org/whl/torch_stable.html

# 2. Install MIM and MMCV
pip install openmim
mim install mmcv-full

# 3. Install SIATune
git clone https://github.com/SIAnalytics/siatune.git
cd siatune
pip install -e '.[optional]'

Hyperparameter tuning with OpenMMLab's model frameworks

Start hyperparameter tuning with existed configuration file

python tools/tune.py ${TUNE_CONFIG} [optional tune arguments] \
    --trainable-args ${TASK_CONFIG} [optional task arguments]

MMDetection

Prepare datasets

Please refer to this link.

Run scripts

# MMDetection Example
mim install mmdet
mim download mmdet --config faster_rcnn_r50_fpn_1x_coco --dest configs/mmdet/faster_rcnn
python tools/tune.py configs/mmdet/mmdet_asynchb_nevergrad_pso.py --trainable-args configs/mmdet/faster_rcnn/faster_rcnn_r50_fpn_1x_coco.py