Shihua Huang

02/2025: Our paper titled " DEIM: DETR with Improved Matching for Fast Convergence" has been accepted to IEEE CVPR 2025. Many thanks to all co-authors for their kind help.
12/2024: Our paper titled " DEIM: DETR with Improved Matching for Fast Convergence" is available at arXiv. DEIM is an advanced training framework that enhances the matching mechanism in DETRs. It sets a new SoTA performance benchmark for real-time object detection without additional data, significantly surpassing YOLOv11.
10/2024: we won 1st place in the Open-set Recognition challenge at the Out Of Distribution Generalization in Computer Vision Workshop, ECCV 2024.
09/2024: Our paper titled " From COCO to COCO-FP: A Deep Dive into Background False Positives for COCO Detectors" is available at arXiv. COCO-FP is a dataset designed to evaluate false positive detections for COCO object detectors.
06/2024: We achieved 1st place in Low-light Object Detection and 3rd place in Instance Segmentation at the Physics-Based Vision meets Deep Learning Workshop, CVPR 2024.
02/2023: Our paper titled " Revisiting Residual Networks for Adversarial Robustness: An Architectural Perspective" is accepted to IEEE CVPR'23. RobustResNets is consisting by the RobustScaling and RobustResblok. Moreover, our RobustResNets achieve new state-of-the-art AutoAttack performance with/without extra data. Many thanks to all co-authors for their kind help.
10/2022: Our papers titled " Multimodal image-to-image translation via a single generative adversarial network" and " Surrogate-assisted Multi-objective Neural Architecture Search for Real-time Semantic Segmentation" both are accepted by IEEE TAI.
12/2021: When paired with our ICCV2021 FaPN, Facebook recent Mask2Former achieved the best performance over ADE20K validation and test set. Moreover, 3 out of top 5 use our FaPN.
07/2021: Our paper titled "FaPN: Feature-aligned Pyramid Network for Dense Image Prediction" is accepted to IEEE ICCV'21. Moreover, our FaPN improved the best MaskFormer for ADE20k-150 by about 1% and achieved 56.7% mIoU, which is the 2nd best among all published methods. Many thanks to all co-authors for their kind help.
07/2021: Our paper titled "RelativeNAS: Relative Neural Architecture Search via Slow-Fast Learning" is accepted to IEEE T-NNLS. Congratulations to Hao Tan for leading the paper.
04/2021: Our paper titled "Efficient Evolutionary Neural ArchitectureSearch by Modular Inheritable Crossover." is accepted to Elsevier SWEVO. Congratulations to Cheng He for leading the paper.
04/2020: Our team (EMI_VR) achieved the best result on Video Deblurring Track of NTIRE 2020 Challenge and our solution is presented as a technical report titled "NTIRE 2020 Challenge on Image and Video Deblurring", which is accepted to IEEE CVPR'20 workshop.
03/2020: Our paper titled "Evolutionary Multi-Objective Optimization Driven by Generative Adversarial Networks (GANs) is accepted to IEEE T-CYB. Congratulations to Cheng He for leading the paper.
1/2020: Our proposal titled "Deep Learning Based Aerofoil Design" is approved by Chinese Ministry of Industry and Information Technology (RMB 3,200,000).
11/2019: Our team (只想划水) won the final champion and received 30K RMB reward at 5th NAVINFO Cup on AutoDriving.
4/2019: Our team (9102) ranked 1st place in the first round over Defense Track at IJCAI-19 Alibaba Adversarial AI Challenge on Defense.
03/2019: A simple and English version of my undergraduate thesis titled "IvaNet: Learning to jointly detect and segment objets with the help of Local Top-Down Modules is released to arXiv. Many thanks for Prof. Lu Wang's supervision.
09/2018: Our paper titled "A local top-down module for object detection with multi-scale features" is accepted to PRCV.