kareem-metwaly / glidenet

This repository contains the implementation and the building blocks of GlideNet and Informed Convolution. This work is published at CVPR 2022 paper titled "GlideNet: Global, Local and Intrinsic based Dense Embedding NETwork for Multi-category Attributes Prediction"

GlideNet: Global, Local and Intrinsic based Dense Embedding NETwork for Multi-category Attributes Prediction

Accepted at The IEEE/CVF Computer Vision and Pattern Recognition - CVPR 2022

This repo contains the implementation of GlideNet and some useful files to reuse its building blocks. This work has been done during my internship at Scale AI. For more information, please check the project's page

Structure of the code

1. models contains different architectures for guidance to reimplement or reuse some building blocks of GlideNet. Most importantly, glidenet.py contains the implemented complete GlideNet structure. In addition, informedconv2d.py contains a PyTorch implementation of the novel Informed Convolution. You can also find some examples of modules that are based on Informed Convolution at informed_resnet.py

2. configs contains examples of configuration files that can be used to define the parameters of GlideNet's architecture.

3. dataset contains PyTorch implementations to retrieve data samples from CAR and VAW datasets after being preprocessed.

4. structures contains some useful abstract classes and dataclasses that were implemented to ease dealing with inputs and outputs of the models and the datasets.

Note: The code has previously used some proprietary packages during my internship at Scale AI. Therefore, these packages are missing here, which breaks the code. However, in models/glidenet.py, you can find the complete implemented structure of GlideNet. You can use configs/models/car/glidenet.yaml for example to play with the configuration of the architecture.

Setting up the environment

All required packages are found in requirements.txt. There are some missing proprietary packages, but they are not essential for building GlideNet and its components.

conda create -n glidenet python=3.8.5
pip install -r requirements.txt

Citation

Please cite our CVPR 2022 paper if you use GlideNet, Informed Convolution or any of the building blocks in your work.

@InProceedings{metwaly_cvpr_2022_glidenet,
    author    = {Metwaly, Kareem and Kim, Aerin and Branson, Elliot and Monga, Vishal},
    title     = {GlideNet: Global, Local and Intrinsic based Dense Embedding NETwork for Multi-category Attributes Prediction},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2022},
}