Implementation of our IEEE/CVF WACVw 2021 paper "PeR-ViS: Person Retrieval in Video Surveillance using Semantic Description". If you find this code useful in your research, consider citing:
author = {Shah, Parshwa and Garg, Arpit and Gajjar, Vandit},
title = {PeR-ViS: Person Retrieval in Video Surveillance Using Semantic Description},
booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) Workshops},
month = {January},
year = {2021},
pages = {41-50}
author = {Galiyawala, Hiren and Shah, Kenil and Gajjar, Vandit and Raval, Mehul S},
title = {Person retrieval in surveillance video using height, color and gender},
booktitle = {Proceedings of the IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS)},
month = {November},
year = {2018},
pages = {1-6}
author = {Galiyawala, Hiren J and Raval, Mehul S and Laddha, Anand},
title = {Person retrieval in surveillance videos using deep soft biometrics},
booktitle = {Proceedings of Deep Biometrics},
month = {January},
year = {2020},
pages = {191-214}
author = {Galiyawala, Hiren and Raval, Mehul S and Dave, Shivansh},
title = {Visual appearance based person retrieval in unconstrained environment videos},
booktitle = {Proceedings of Image and Vision Computing},
month = {December},
year = {2019},
pages = {103816}
author = {Shah, Priyansh and Raval, Mehul S and Pandya, Shvetal and Chaudhary, Sanjay and Laddha, Anand and Galiyawala, Hiren},
title = {Description based person identification: use of clothes color and type},
booktitle = {Proceedings of the National Conference on Computer Vision, Pattern Recognition, Image Processing, and Graphics},
month = {April},
year = {2018},
pages = {457-469}
This code was initially tested on an Ubuntu 16.04 system using Keras 2.0.8 with Tensorflow 1.8 backend.
Please open and follow the instruction under PeR-ViS WACVw 2021 folder. PeR-ViS Track Unpublished is the newer version considering a more complex solution by adding tracker in the framework, it also uses YOLO for better FPS in real-time. More updates will be made available after pre-print.