Skip to content

Latest commit

 

History

History
59 lines (39 loc) · 2.26 KB

README.md

File metadata and controls

59 lines (39 loc) · 2.26 KB
Raw

[CVPRW] 2023 AI City Challenge: Tracked-Vehicle Retrieval by Natural Language Descriptions With Multi-Contextual Adaptive Knowledge

🏆 The 1st Place Solution to The 7th NVIDIA AI City Challenge (2023) Track 2: Tracked-Vehicle Retrieval by Natural Language Descriptions.

[official results] [paper] [slides]

image

Framework

image

The configuration files are in src/configs.py. Set up the right data path first: DATA_DIR - path to store dataset folder (extracted frames)

ROOT_DIR - path to source code folder

Pre-Processing

Download the model checkpoints for srl in folder pre_processhere and place it in folder src/pre_process/weight/

1. bash scripts/prepare_pre_process.sh
2. bash scripts/pre_process.sh

Training

bash scripts/train.sh

Inference

Change the INFERENCE_FROM in your configuration file to the checkpoints downloaded in folder recognition and retrieval here, and run the below bash scripts to extract embeddings and ranking the query videos.

bash scripts/inference.sh

Post-Processing & Submission

bash scripts/post_process.sh

Others

If you have any questions, please leave an issue or contact us: [email protected].

@InProceedings{Le_2023_CVPR,
    author    = {Le, Huy Dinh-Anh and Nguyen, Quang Qui-Vinh and Luu, Duc Trung and Chau, Truc Thi-Thanh and Chung, Nhat Minh and Ha, Synh Viet-Uyen},
    title     = {Tracked-Vehicle Retrieval by Natural Language Descriptions With Multi-Contextual Adaptive Knowledge},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
    month     = {June},
    year      = {2023},
}