README.md

hpc_yolo3

Introduction

A keras (tensorflow backend) implement of YOLOV3 (2D image object detection) for UNT REU2019 PROJECT.

Created by Xu Ma, xuma@my.unt.edu.

Requirement

UNT HPC environment, which includes python3, tensorflow, keras, numpy ...

Examples

Install

1. Login UNT HPC sever: ssh username@talon3.hpc.unt.edu.

2. Create a new folder using your name. For example, mkdir Jacob .

3. cd Jacob.

4. Download hpc_yolo3 project.

   git clone https://github.com/13952522076/hpc_yolo3.git
   

   Make sure you download the project in your name folder.

5. Download weight to local computer: yolo.h5

   Copy the downloaded yolo.h5 file to hpc sever 'Jacob/hpc_yolo3/model_data' folder using scp.

   (Only this step is in a new terminal.)

   scp /download/yolo.h5 username@talon3.hpc.unt.edu:/home/username/Jacob/hpc_yolo3/model_data
   

   Remember 1). change the local path to yolo.h5 file, and change the 2) username (2 times), 3)Jacob.

   3 modifications on this command.

6. You can have a look at Images folder, which includes 3 subfoloder.

   images: the collected image data.

   outputs: the detected labels, confidence, coordinate

   results: resulted detection images.

7. Remove all files in these three folders. Do not delete the folders.

    rm Images/results/*
    rm Images/outputs/*
    rm Images/images/*
   

   upload collected images to 'Images/images/' folder.

   Edit 'run_test.sh', change demo@my.unt.edu to your email, username to hpc username, YOURNAME to your name.

   Run SLURM job 'run_test.sh' by

   sbatch run_test.sh
   

   After this, you will see your job id, like 999444. A few seconds later, a log file will generated, named as hpc_999444.log.

   Once failed or completed, you will get an email.

   Detected images will be in Images/results. Detected information will be in Images/outputs.