University College London
5 files

Coin Reconstruction Data

posted on 2024-01-04, 16:01 authored by Mark AltaweelMark Altaweel, Adel Kehlifi, Mohammad Hashir Zafar

These data are used for training, testing, and evaluating the CycleGAN project related to coin reconstructions. Data are used to train the given CycleGAN model and coins are used to evaluate the model both internally, computationally, and by other users. This file provides training data obtained used to train the CycleGAN. There is a training folder which contains two folders (badPNG and goodPNG). The goodPNG folder contains coins that are well preserved and used to train what a coin should look like. The badPNG folder contains damaged coins and attempts to train to reconstruct from these coins. There are two metadata files (bad coins metadata and good coin metadata) provide information about the coins and where they are found through the links provided in the files. These files correspond to the well preserved coins (good coin data) and poorly preserved coins (bad coins) used for reconstruction. These are image data used to test the developed CycleGAN model. The images folder contains the testing data. An index file (HTML) indicates the coins present. The index and images indicate real (i.e., the real coins used in reconstruction) and fake (i.e., reconstructed or GAN-derived coins).

CoinsEvaluation.pdf: This file is used by testers to evaluate if they think given reconstructed coins shows improved quality or not from the real coin. The users are asked to evaluate coins on a 1-5 scale, with 1 indicating no difference and 5 indicating much improved reconstructions. This folder contains real and fake (i.e., reconstructed coins) that is used to ask judges or users to determine if they can tell if a given coin is real or reconstructed (i.e., fake). In addition to the real and reconstructed coins, a pdf (Real_or_Fake.pdf) is used to ask judges to determine which given coins are real and which are generated reconstructions. This is the final CycleGAN model used for reconstructions applied in the paper.


Usage metrics

    University College London



    Ref. manager