Skip to content

Latest commit

 

History

History
50 lines (42 loc) · 1.31 KB

README.md

File metadata and controls

50 lines (42 loc) · 1.31 KB

Scene Graph Generation Challenge

DagsHub x ML@Purdue Hackathon

This is the template repository for the Scene Graph Generation Challenge.

Setup

From the root of the repository:

pip install -r requirements.txt  # please use a virtual environment!
aws s3 cp s3://visualgenome data --recursive --no-sign-request

Forking the template to your repository

git remote add base https://dagshub.com/ML-Purdue/sgg-template.git
git pull base --rebase
git branch -M main
git push origin main

Updating the Fork

git pull base main --allow-unrelated-histories

Then, resolve any merge conflicts, commit and push!

Inference Pipeline

python -m src.model.arch

Project Structure

├── LICENSE
├── README.md
├── requirements.txt
├── data                    <- visual genome dataset
└── src
    ├── const.py            <- constants (hyperparameters, paths)
    ├── data
    │   └── visualgenome.py <- dataset sampler
    ├── model
    │   ├── arch.py         <- model architecture
    │   ├── agcn.py         <- attention graph convolutional network
    │   └── repn.py         <- relationship proposal network
    └── utils.py            <- helper functions (nms)