layout-eval

Setup

We use poetry to manage dependencies. Install poetry and run the following command to install dependencies.

poetry install

Data preparation

Download the pre-processed dataset and generated layouts by running the following command.

wget https://github.com/mayu-ot/ltsim/releases/download/v1.0.0-alpha/data.zip
unzip data.zip

The data directory should look like this:

data
├── datasets # post-processed datasets
│   ├── rico25
│   │   ├── test.json
│   │   ├── train.json
│   │   └── val.json
│   └── publaynet
├──fid_feat # pre-extracted features for FID evaluation
├── results_conditional # generated layouts for conditional layout generation
│   ├── publaynet
│   └── rico
└── results_conditional # generated layouts for unconditional layout generation
    ├── publaynet
    └── rico
        ├── partial # generated layouts for layout completion
        └── c # generated layouts for label-conditioned layout generation
            ├── bart
            ├── ...
            └──vqdiffusion

FID feature extraction

Download the LayoutDM resources and copy download/fid_weights/FIDNetV3/rico25-max25/model_best.pth.tar to $FID_WEIGHT_FILE.
Run the following command to extract layout features to evaluate FID on RICO dataset.

python src/experiments/feature_extraction.py \
  --dataset_type rico25 \
  --input_dataset_json $DATASET_JSON \
  --output_feat_file $OUTPUT_FILE_NAME \
  --fid_weight_file $FID_WEIGHT_FILE

Evaluate conditional layout generation

Download generated layouts in ./data following the instruction. Run the script to get evaluation results on RICO. The results are saved in data/results/eval_conditional/rico/result.csv

poetry run python src/experiments/eval_conditional.py rico

Evaluate unconditional layout generation

Download generated layouts in ./data following the instruction. Run the script to get evaluation results on RICO. The results are saved in $RESULT_FILE.

poetry run python src/experiments/eval_unconditional.py rico $RESULT_FILE

Demo

To run an iteractive app to try the evaluation metrics, run the following command.

streamlit run src/app/measure_explore.py