Shilong Zhang

Shilong Zhang(张士龙)

Shilong, presently in his second year (2023-present) as a Ph.D. student at the Department of Computer Science, The University of Hong Kong (HKU), is being mentored by Prof. Ping Luo. In his previous stint, he was an integral part of the team headed by Kai Chen while serving as a core developer for MMDetection and MMCV.

He completed his Bachelor's degree from the University of Science and Technology of China (USTC) in 2019, distinguishing himself as one of the top 4% outstanding graduates.

At present, Shilong's research is primarily focused on building efficient vision generative models and algorithms. He welcomes intellectual conversations about these areas of study.

Email / Google Scholar / Github

Recent News

[2025/2/10] We present FlashVideo, an efficient paradigm for text-to-video generation.
[2024/3/26] We propose FlashFace that can generate high ID fidelity images in seconds.
[2023/7/7] We present a vision and language model named GPT4RoI to do region-level image understanding.
[2023/4/26] We present a vision and language model named MultiModal-GPT .
[2023/3/20] Two papers was accepted by CVPR 2023. DDQ DETR achieve 52.1 AP with R-50 backbone within 12 epochs.
[2022/3/15] One paper was accepted by CVPR 2022.
[2021/11/27] We release MMFewShot, an open source few shot learning toolbox based on PyTorch.
[2021/5/8] One paper was accepted by ICML 2021.
[2020/2/24] One paper was accepted by CVPR 2020.
[2019/6/28] Awarded as outstanding graduates by USTC.

Publications

[1] FlashVideo: Flowing Fidelity to Detail for Efficient High-Resolution Video Generation
Shilong Zhang*, Webo Li*, Shoufa Chen, Chongjian Ge, Peize Sun, Yida Zhang, Yi Jiang, Zehuan Yuan, Binyue Peng, Ping Luo
(a) Dividing the process into prompt fidelity and quality enhancement stages, delivering a stunning reduction in DiT's computational load . (b) Enableing users to preview the initial output and accordingly adjust the prompt before committing to full-resolution generation, thereby significantly reducing wait times and enhancing commercial viability.
Code has been released at this repo !

[2] FlashFace: Human Image Personalization with High-fidelity Identity Preservation
Shilong Zhang, Lianghua Huang, Xi Chen, Yifei Zhang, Zhi-Fan Wu, Yutong Feng, Wei Wang, Yujun Shen, Yu Liu, Ping Luo
This work presents FlashFace, a practical tool with which users can easily personalize their own photos on the fly by providing one or a few reference face images and a text prompt
Code has been released at this repo !

[3] GPT4RoI: Instruction Tuning Large Language Model on Region-of-Interest Shilong Zhang*, Peize Sun*, Shoufa Chen, Min Xiao, Wenqi Shao, Wenwei Zhang, Kai Chen, Ping Luo
(* Equal contribution) ,
We present a vision and language model named GPT4RoI to do region-level image understanding.
Code has been released at this repo !

[4] MultiModal-GPT: A Vision and Language Model for Dialogue with Humans
Tao Gong*, Chengqi Lyu*, Shilong Zhang*, Yudong Wang*, Miao Zheng*, Qian Zhao*, Kuikun Liu*, Wenwei Zhang*, Ping Luo, Kai Chen
(* random order) ,
We present a vision and language model named MultiModal-GPT to conduct multi-round dialogue with humans.
Code has been released at this repo !

	[5] Dense Distinct Query for End-to-End Object Detection Shilong Zhang, Xinjiang Wang, Jiaqi Wang, Jiangmiao Pang, Chengqi Lyu, Wenwei Zhang, Ping Luo, Kai Chen CVPR2023( Equal contribution)* , DDQ-DETR achieves 52.1 AP on MS-COCO dataset within 12 epochs using a ResNet-50 backbone, outperforming all existing detectors in the same setting. Code has been released at this repo !
	[6] Consistent-Teacher: Towards Reducing Inconsistent Pseudo-targets in Semi-supervised Object Detection Xinjiang Wang, Xingyi Yang, Shilong Zhang, Yijiang Li, Litong Feng, Shijie Fang, Chengqi Lyu, Kai Chen, Wayne Zhang CVPR2023 ( Equal contribution)* , It achieves 40.0 mAP with ResNet-50 backbone given only 10% of annotated MS-COCO data, which surpasses previous baselines using pseudo labels by around 3 mAP. When trained on fully annotated MS-COCO with additional unlabeled data, the performance further increases to 47.2 mAP. Code has been released !
	[7] Group R-CNN for Point-based Weakly Semi-supervised Object Detection Shilong Zhang, Zhuoran Yu, Liyang Liu, Xinjiang Wang, Aojun Zhou, Kai Chen CVPR2022 (* Equal contribution)* , We study the problem of weakly semi-supervised object detection with points (WSSOD-P). Group R-CNN significantly outperforms the prior method Point DETR by 3.9 mAP with 5% well-labeled images. Code has been released !
	[8] Group Fisher Pruning for Practical Network Compression Liyang Liu, Shilong Zhang,Zhanghui Kuang,Jing-Hao Xue ,Aojun Zhou ICML2021 ( Equal contribution)* , We present a general channel pruning framework for complicated structures ! Code has been released !
	[9] Scale-equalizing Pyramid Convolution for object detection Xinjiang Wang, Shilong Zhang , Zhuoran Yu, Litong Zhang, Wayne Zhang CVPR2020 ( Equal contribution)*, We proposed a scale-equalizing pyramid convolution method that relaxes the discrepancy between the feature pyramid and the gaussian pyramid. The module boosts the performance about 3.5 mAP in single-stage object detection with negligible inference time. Code has been released !

Stolen from Jon Barron