README.md

Naic-preliminary-contest

Introduction

Rank 3th in the NAIC 2020(National Artificial Intelligence Challenge) preliminary contest.

Setting

• RTX2080Ti × 4
• Linux
• Cuda10.1

Usage

1. Install packages

• The code is tested on python3.5 and torch 1.1.0. (A higher version should be also work)

pip install -r requirements.txt  

2. Dataset

• Download the dataset on NAIC
• Please put dataset: train, image_A, image_B in folder data

3. Train

• Run this command at the terminal:

sh autorun_train.sh

• You can change the python version in autorun_train.sh

4. Test

• The weights is available on line:
  百度网盘：https://pan.baidu.com/s/13efsRZx7qdqVCX5cJSf1-g
  提取码：s7f2
  Please download the weights and put it in folder weights
• Testing command:

sh autorun_test.sh

5. Ensemble

• The best results in rank A and rank B, you can change the test path in folder dataset to get the results of dataset image_A and image_B.

sh autorun_ensemble.sh

6. Results

• Stored in folder results

sh zip.sh

• Compressed to ZIP format.