Qiming Huang

I'm Qiming Huang (黄麒铭), a final-year PhD student at MIx Group @ University of Birmingham, supervised by Prof. Jianbo Jiao. I obtained my B.Eng. degree in Data Science at Xi'an Jiaotong–Liverpool University in 2023.

Email / Scholar

News

2026-11: Will visit VGG@Oxford University to work with Prof. João F. Henriques on robotic planning.
2026-09: One paper about multimodal attribution is accepted to 2026! See you in Australia! 🦘 🇦🇺
2026-07: Will join Durham University as Visiting Researcher with Prof. Hubert P. H. Shum.
2026-05: Glad to be recognised as an ICML 2026 Gold Reviewer.
2025-11: One paper about training-free open-vocabulary semantic segmentation is accepted to WACV2026. See you in Tucson! 🌵 🇺🇸
2025-09: Will start internship at Samsung R&D Institute UK (SRUK).
2025-07: Two papers are accepted to BMVC, congrats to the co-authors!
2025-05: We are organising the BinEgo-360 Workshop and Challenge at ICCV 2025 in Hawaii. Please participate in the Challenge to get various awards, and join us to present your papers! See you in Hawaii! 🌺 🇺🇸
2025-01: One paper is accepted to ICLR2025! See you in Singapore 🇸🇬
2024-10: Will be serving as volunteer for ECCV 2024. See you in Milan 🇮🇹
2024-09: One paper is accepted to ACCV2024, congrats to the co-authors. See you in Hanoi 🇻🇳
2023-09: Starting my PhD at the School of Computer Science, University of Birmingham.

Research

I'm interested in open world learning problems in computer vision, focusing on the following aspects: 1) Open-world visual understanding, open vocabulary task 2) Visual Reasoning. If you are interested in open world learning, you are more than welcome to drop me an email. I am eager to brainstorm ideas related to open world learning.

	TIB: Sample-wise Tempered Information Bottleneck for Multimodal Attribution beyond Alignment Assumption Qiming Huang, Peixi Liu, Jianbo Jiao NeurIPS 2026 TIB adapts information bottleneck attribution to each image-text pair's reliability, avoiding forced explanations when the modalities are semantically misaligned.
	Structure-Aware Feature Rectification with Region Adjacency Graphs for Training-Free Open-Vocabulary Semantic Segmentation Qiming Huang, Hao Ai, Jianbo Jiao WACV 2026 We use image-derived region adjacency graphs to refine CLIP features for training-free open-vocabulary semantic segmentation, reducing local noise and improving region consistency.
	Revisit the Open Nature of Open Vocabulary Semantic Segmentation Qiming Huang, Han Hu, Jianbo Jiao ICLR 2025 code / paper We introduce a mask-wise evaluation protocol that accounts for semantic ambiguity in open-vocabulary segmentation.
	Stepwise Feature Fusion: Local Guides Global Jinfeng Wang, Qiming Huang, Feilong Tang, Jia Meng, Jionglong Su, Sifan Song MICCAI 2022* code / pdf Simple but powerful baseline for medical image segmentation.