Unified Bee Runner is a pipeline for processing and analyzing bee-related datasets. This project includes several steps such as dataset creation, video conversion, data splitting, and model training. The model is meant to run on ilab, and through slurm, and for the Behavioral Analysis project at Rutgers WINLAB.
This is mean to unify all the code that has been created for the project (2024 presentation attached).
-
Clone the repository:
git clone https://github.com/Elias2660/Unified-bee-Runner.git
If there is a branch that you want to use, you can simply checkout to the branch that you choose.
The most easy way to run the runner is to run it directly in the directory that contains the data. After cloning, your directory should look something like this:
data_dir
├── Unified_bee_Runner
├── d1.mp4
├── d2.mp4
├── d3.mp4
├── logNo.txt
└── logPos.txt
To run with default settings, you can run the Slurm_Run.sh file, which has preset sbatch settings (this is in the data_dir):
./Unified-bee-Runner/Slurm_Run.shRun squeue -u <user> to be able to find your current jobs and the servers that they are running on. If you don't see that job, try again and if it doesn't work again, the check the slurm troubleshooting guide.
For more advanced usage, you can checkout the Unifier_Run.sh file, which contains most of the settings that are used in the pipeline. To see a more advanced list, you can also check out the ArgParser.py file that has all the possible settings that can be used. Not all of them work and this pipeline is still working through many bugs.
The most important settings that you can change are:
--number-of-samples: the max number of samples that the sampler will aim to extract from each video--max-workers-video-sampling: the number of concurrent videos that you can sample with--frames-per-sample: the number of frames that will be used by the model for each sample for concurrence.
Additionally, the INPUT_PATH and OUTPUT_PATH variables, both in Unifier_Run.sh can be edited to change the where the runner will look for input and output the experiment results.
Changing the slurm settings can be helpful. Updating memory could be dangerous unless you are knowledgeable of the workflow, but using the -x and an ilab server that seems full could help your job not get canceled.
This pipeline automatically generates a config.log file inside the Unified-bee-Runner dir. The pipeline automatically looks inside the dir in order to find the file, but if it doesn't it uses the arguments passed in the Unifier_Run.sh file. You can copy the config file into the Unified-bee-Runner across runs in order to use the same setting.
If you slurm job keeps dying, then the easiest way to fix that problem is the reduce memory, which can most often be done by reducing --max-workers-video-sampling argument in Unifier_Run.sh and the --mem= command in Slurm_Run.sh. Usually you can reduce the argument to around 500GB without changing the --max-workers-video-sampling argument. Otherwise, you should reduce by one for the --max-workers-video-sampling argument for each ~100 GB you remove. You can also remove / decrease the -n, -c, -G arguments, but that might slow stuff down.
This is run using the chapter system, so you can choose the specific steps that are used here by editing the --start and --end commands, which are by default respectively set at 0 and 6, which will run the entire model.
Video Conversion and Counting: Converts .h264 videos to .mp4 format and creates counts.csv.Background Subtraction: Applies background subtraction to the video frames to isolate the beesDataset Creation: Creates the dataset.csvData Splitting: Splits the data into training and testing setsVideo Sampling: Clones the VideoSamplerRewrite repository and samples the video framesModel Training: Runs the model training script
You can choose specific cropping arguments by editing the --crop, crop_x_offset, crop_y_offset, --out-width, and --out-height.
The initial offset coords are in the center of the image.
Y offset: Negative makes it higher while positive makes it lower
X offset: Positive moves it right while negative moves it left
Out-width: The width of the cropped image
Out-height: The height of the cropped image
This projects has other project submodules in the non_workflow_links directory, which relate to other parts of the bee project.
Contributions are welcome! Please open an issue or submit a pull request.
Please note that this project is released with a Contributor Code of Conduct. By participating in this project you agree to abide by its terms.
If you discover a security vulnerability within this project, please report it. We will respond to your report promptly and work with you to understand and address the issue.
This project is licensed under the MIT License.