This is an unofficial implementation of RegNet (Designing Network Design Spaces).
(TODO : requirements.txt and Dockerfile for the image of fixed environment.)
- python >= 3.6
- tensorflow >= 2.2
- First, set the design space in
config.yml
- Second, sample models in the design space
>>> python search_space.py --config /path/of/config.yml --num_model $NUM_MODEL --model_name $MODEL_NAME
or
>>> ./search.sh
- Then, train models which are already set
>>> ./main.sh
(TODO : shell command for evluation)