CVPR 2021 Challenge

Track 5: Natural Language-Based Vehicle Retrieval

Overview

• This is the code for the #4 solution [paper]

Prerequisites

• torch
• torchvision
• numpy
• pandas
• sklearn
• opencv-python
• efficientnet_pytorch
• transformers
• albumentations
• tqdm
• timm
• textdistance
• openai/CLIP

Running setup.sh also installs the dependencies

Reproceduring

• Running run_nce.py with the default arguments

Citation

@inproceedings{nguyen2021contrastive,
  title={Contrastive learning for natural language-based vehicle retrieval},
  author={Nguyen, Tam Minh and Pham, Quang Huu and Doan, Linh Bao and Trinh, Hoang Viet and Nguyen, Viet-Anh and Phan, Viet-Hoang},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={4245--4252},
  year={2021}
}