VISUM 2022 Project

Team Beach Nerds' repository of the Project of the VISUM Summer School 2022.

Info

Check out our branch tsar to see our proposed method! The main branch is only the baseline with our improvements.

Create a SSH key for GitHub

First, you should create a SSH key for GitHub. You can go to the official GitHub tutorial here.

Clone repository

Then, to clone this repository you have to open a Terminal and run the following command:

$ git clone git@github.com:visum-summerschool/visum-competition2022.git

Download database

Before training the models, we must be sure that the database structure is built. To download the data and build the structure, run the following command.

$ cd visum-competition2022
$ bash download_data_participants.sh

Create a Virtual Environment and Activate It

$ python -m venv venv
$ source venv/bin/activate

Install the requirements to set up your Python environment

To intall the minimum requirements you just need to run the following

$ pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113
$ pip install -U albumentations
$ pip install torchinfo
$ pip install tqdm
$ pip install tensorboard
$ pip install -U cython
$ pip install pycocotools

Train model

To train the baseline model, you have to run the following command (please be sure that your current directory is the root directory of the repository):

$ python code/model_train.py

This command has several constant variables you can change:

BATCH_SIZE - the batch size for the DataLoader
NUM_EPOCHS - the number of epochs for the training
IMG_SIZE - the image size you will use (H, W)
VAL_MAP_FREQ - the frequency you want the training loop to print the mAP values

Creating and changing your own models

To do this, you can go to model_utilities.py and edit the LoggiBarcodeDetectionModel.

Plot results

If you want to check and visualise some of you results on validation (or training) you can run the following command:

$ python code/plot_results.py

Creating a submission

After the training of your models you have to create a submission file. This file will be .ZIP file containing: 1) the code/ directory, the results/models/visum2022.pt directory, and the Dockerfile. Be sure that you fill the requirement needs of your Dockerfile, and the run the following command:

$ bash create_submission_file.sh

DubiousCactus/beach_nerds