EAST-for-3D-Digital-Rocks

This repository is an official PyTorch implementation of the paper "Efficiently Reconstructing High-Quality Details of 3D Digital Rocks with Super-Resolution Transformer ".

The source code is primarily derived from EDSR. We provide full training and testing codes. You can train your model from scratch, or use a pre-trained model to enlarge your digital rock images. We will upload the pre-trained model soon.

Code

Dependencies

• Python 3.8.5
• PyTorch = 2.0.1
• numpy
• cv2
• skimage
• tqdm

Quick Start

git clone ### Quick Start

```bash
git clone https://github.com/MHDXing/MASR-for-Digital-Rock-Images.git
cd EAST-for-3D-Digital-Rocks-main/src

Dataset

The dataset we used was derived from DeepRockSR-3D. There are 2400, 300, 300 HR 3D images (100x100x100) for training, testing and validation, respectively.

Training

1. Download the dataset and unpack them to any place you want. Then, change the dir_data argument in ./options.py or demo.sh to the place where images are located
2. You can change the hyperparameters of different models by modifying the files in the ./options.py
3. Run main.py using script file demo.sh

bash demo.sh

4. You can find the results in ./experiments/EAST if the save argument in ./options is EAST.

Testing

1. Download our pre-trained models to ./models folder or use your pre-trained models
2. Change the dir_data argument in ./options.py or demo.sh to the place where images are located
3. Run main.py using script file demo.sh

bash demo.sh

4. You can find the enlarged images in ./experiments/results folder.