Pearl

This repository contains the code for Pearl.

Installation

Create a Python 3.10 virtual environment using pyenv and install dependencies:

pyenv virtualenv 3.10.0 pearl
pyenv local pearl # cuda=11.8

Data Layout

Before training or evaluation, make sure the datasets are already placed under data/ with the following structure:

The dataset can be downloaded from: https://huggingface.co/datasets/hiranohachiman/Spica

data/
├── images/
├── spica_test.csv
├── spica_train.csv
└── spica_val.csv

Feature Extraction

Next, generate the features by following the instructions described in save_features/README.md. The commands there will populate the required feature files under save_features/.

setup poetry

poetry sync

Training

sh train.sh

Evaluation

sh validate.sh