PaperCodeReview / RegNet-TF

TF 2.x implementation of RegNet (Designing Network Design Spaces, CVPR 2020).

RegNet-TF

This is an unofficial implementation of RegNet (Designing Network Design Spaces).

Requirements

(TODO : requirements.txt and Dockerfile for the image of fixed environment.)

• python >= 3.6
• tensorflow >= 2.2

Training

Set design space

• 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

Evaluation

(TODO : shell command for evluation)