VehicleFinder-CTIM: a keyword-based text-image cross-modal vehicle retrieval system

VehicleFinder

Since the whole system would be used for commercial purposes, we only open-source the core module CTIM (contrastive text-image module).

NanoDet: link | Dataset-> UA-DETRAC: link password: bygu

BiLSTM-CRF: link | Dataset-> FindVehicle: link

---

CTIM

Requirements:

einops==0.4.1
gensim==4.1.2
jieba==0.42.1  
matplotlib==3.5.2   
numpy==1.22.4+mkl 
opencv_python_headless==4.5.5.64 
pandas==1.4.2 
Pillow==9.2.0
scipy==1.8.1
thop==0.1.0.post2206102148
torch==1.9.1+cu111
torchvision==0.10.1+cu111 
tqdm==4.64.0

Dataset

[multi-label -> vehicle proposal] cross modal matching Baidu Cloud Disk password: 8iqk

Forward from [UA-DETRAC-ML](https://github.com/GuanRunwei/UA-DETRAC-ML), please cite it if you use it in your research

{
@misc{uadetracml,
title={UA-DETRAC-ML},
author={Runwei Guan},
howpublished = {\url{https://github.com/GuanRunwei/UA-DETRAC-ML}},
year = {2022},
}

Implementation

conda create -n CTIM
conda activate CTIM
pip install requirements.txt 
python train.py

The code of this project is clear, you could find out and replace the hyperparameters and file paths without any difficulty.