A keras (tensorflow backend) implement of YOLOV3 (2D image object detection) for UNT REU2019 PROJECT.
Created by Xu Ma, xuma@my.unt.edu.
UNT HPC environment, which includes python3, tensorflow, keras, numpy ...
-
Login UNT HPC sever:
ssh username@talon3.hpc.unt.edu
. -
Create a new folder using your name. For example,
mkdir Jacob
. -
cd Jacob
. -
Download hpc_yolo3 project.
git clone https://github.com/13952522076/hpc_yolo3.git
Make sure you download the project in your name folder.
-
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.
-
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.
-
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.