xziyue/SciEye

SciEye: A System for Finding the Underlying Datasets for Scientific Figures

Authors: Ziyue "Alan" Xiang, Edward J. Delp

Direct all correspondence to Edward J. Delp, ace@ecn.purdue.edu.

Installation and Configuration

• OS: Ubuntu 20.04

• Set up conda environment

  • conda create -n data-graph-matching

  • conda env update -n data-graph-matching --file environment.yaml

• Acquire the JSON service key from Google Cloud Platform and save it in /notebook as api_key.json (tutorial)

• Download model checkpoints from https://darknet.ecn.purdue.edu/~xiang71/scieye/scieye_ckpt_v01.zip and extract them to /ckpt

Demo

Please see /notebook/demo.ipynb

Parameters

The parameters of many steps are defined in /data_graph_matching/param.py.

• parallel_n_jobs controls the number of parallel jobs; setting it to 1 can be beneficial for debugging

Dataset

The dataset used to train the Mask RCNN can be found here.