Bidirectional Layout Transformer (BLT)

BLT: Bidirectional Layout Transformer for Controllable Layout Generation In ECCV'22.

원본 코드 출처: layout-blt

Introduction

Automatic generation of such layouts is important as we seek scale-able and diverse visual designs. We introduce BLT, a bidirectional layout transformer. BLT differs from autoregressive decoding as it first generates a draft layout that satisfies the user inputs and then refines the layout iteratively.

Set up environment

conda env create -f environment.yml
conda activate layout_blt

pip install jaxlib==0.1.69+cuda110 -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html

출처로 남긴 원본 코드에서 환경 구축 시 유의 사항

원본 코드에서 environment.yml이 update되지 않았다면 아래 링크를 참고해 jax.Array가 jax.numpy.DeviceArray 로 변경되어 있는지 확인해주세요.

• [commit error] layout-blt/utils/layout_bert_fast_decode.py
• [commit error] layout-blt/utils/layout_fast_decode.py

Running

# model_dir 이름 형식: 실험 번호와 실험 이름을 '_'로 구분
# 예시) exp1_expName
# Training a model
python  main.py --config configs/${config} --workdir ${model_dir}

# Testing a model
python  main.py --config configs/${config} --workdir ${model_dir} --mode test

# Evaluating a model with IOU
python main.py --config configs/${config} --workdir ${model_dir} --mode eval