Chenyuan Qu

Computer vision, multimodal learning, and software systems.

Chenyuan Qu

I am a PhD student at the University of Birmingham and Head of Technologies at Allsee and Vieunite.

My research focuses on computer vision, multimodal learning, and generative models. Alongside my doctoral work, I work on backend services, internal software, and applied machine-learning systems at Allsee and Vieunite.

Google ScholarGitHubORCIDEmail

University of Birmingham · Allsee · Vieunite

Portrait of Chenyuan Qu

Research interests

01

Interpretable image representations

Separating visual information into structures such as edges, colour regions, and intensity.

VisualSplit02

Multimodal scene understanding

Learning across viewpoints, images, audio, text, location, and spatial context.

360+x03

Generative computer vision

Using generative models for representation learning, reconstruction, restoration, and editing.

VisualSplit

Research

Selected research

Two recent projects on interpretable image representation and multimodal scene understanding. The complete publication list follows below.

BMVC · First-author research2025

BMVC 2025 · Image representation

VisualSplit

VisualSplit studies image representations based on three classical visual descriptors: edges, colour segmentation, and grey-level histograms.

Takeaway

Separating geometry, colour, and illumination into explicit descriptors provides an interpretable representation that can also be edited at the descriptor level.

Question
Learned image features are often difficult to interpret because geometry, colour, and illumination information are represented together.
My role
In this first-author work, we use the three descriptors as separate inputs for image reconstruction and examine their use in editing, restoration, and diffusion-guided generation.
Publication
The work was published at BMVC 2025. The paper, supplementary material, presentation, code, model weights, and examples are publicly available.
ProjectPaperCodeModels
CVPR Oral · Collaborative research2024

CVPR 2024 · Oral paper

360+x

360+x is a dataset for studying scene understanding across multiple viewpoints and aligned sensory modalities.

Takeaway

Aligning panoramic, frontal, and egocentric views with audio, location, and text supports scene-understanding research beyond single-view recognition.

Question
Many scene-understanding datasets focus on a single camera view or modality. 360+x records panoramic, frontal, and egocentric views together with spatial, audio, location, and textual signals.
My role
I was one of six authors on the project. The dataset, benchmark resources, and publication were produced collaboratively by the research team.
Publication
The work was selected for an oral presentation at CVPR 2024. The project site provides the paper, supplementary material, code, dataset access, poster, and teaser video.
ProjectPaperCodeDataset

Research output

Publications

Peer-reviewed papers with links to the paper, code, datasets, citation text, and available project media.

2025
2024
2023

Datasets

2025

BinEgo-360

A binocular egocentric and 360° panoramic multimodal dataset and challenge surface for scene understanding, aligned with spatial audio, text, and geo-metadata.

ChallengeRepository

2026

text-to-art-database

A privacy-safe text-to-image dataset released on Hugging Face, repacked into Parquet shards with embedded image bytes and organised into samples and iterations splits.

Dataset

Experience

Experience & education

Research appointments, industry roles, and education.

Research appointments

University of Birmingham · MI X Group

Feb 2023 — Present
PhD Student
2023 — Present

Doctoral research in computer vision, multimodal learning, and generative AI within the School of Computer Science.

Research Assistant
Dec 2023 — Present

Research on compositionality for foundation models, with a focus on representation, reasoning, and generative computer-vision systems.

Research Assistant
Feb 2023 — Dec 2023

Research on interpretable hydrological modelling and machine-learning methods for scientific analysis.

Education

2021 — 2022

Postgraduate study completed at Birmingham before beginning doctoral research.

2018 — 2021

Undergraduate training in physics, which continues to shape how I think about machine learning and computer vision.

Industry experience

Allsee · Vieunite

Sep 2022 — Present
Head of Technologies
Dec 2024 — Present

Responsible for technology planning across backend architecture, internal software, AI development, and operational systems.

Details4 items
  • Coordinate technical priorities across hardware, software, and internal workflows, with attention to maintainability and delivery requirements.
  • Use Linear to organise requests from sales, product, operations, and engineering into roadmaps, assigned work, and progress tracking.
  • Evaluate and implement machine-learning, generative-AI, and automation use cases in internal and product-facing workflows.
  • Translate requirements discussed with leadership, operations, and sales into technical designs and implementation plans.
Full-stack Engineer
Dec 2023 — Dec 2024

Worked on the CMS and ERP platforms, including consolidating board-specific CMS code into a shared architecture for multiple hardware boards and operating systems.

Details3 items
  • Refactored the CMS so a shared codebase could support multiple boards, deployment targets, and system environments.
  • Developed the internal ERP system, including its core workflows, data structures, and interfaces.
  • Worked across frontend, backend, deployment, and device constraints in response to product requirements.
Algorithm Engineer
Sep 2022 — Dec 2023

Worked on backend services, recommendation components, and Vieutopia AI art features for Allsee and Vieunite.

Details3 items
  • Implemented backend services for AI-related product features and internal workflows.
  • Developed recommendation components for content and product discovery.
  • Developed Vieutopia AI art functionality using generative models.

AsiaInfo Software Co. Ltd

Jul 2020 — Sep 2020
Algorithm Engineer Intern
Jul 2020 — Sep 2020

Developed and optimised machine-learning components for a visual customer-service anchor in a mobile deployment setting.

News

Recent updates

Publications, presentations, and research updates.

Started Help To Grow: Management at BCU

I started the 12-week Help To Grow: Management Course at Birmingham City University Business School, with sessions spanning strategy, digital transformation, marketing, operations, finance, and growth planning.

Source

VisualSplit accepted to BMVC 2025

Project page, paper, supplementary material, code, and model weights are publicly available.

Source

DIFF presented at ICASSP 2025

Collaborative work on diffusion features for cross-domain semantic segmentation.

Source
Earlier news3 entries

360+x selected for a CVPR 2024 oral presentation

Dataset paper with accompanying benchmark resources, code, and public dataset access.

Source

Started my PhD in the MI X group

I began doctoral research in 2023 on computer vision and multimodal learning in the MI X Group.

Source

MeD published at ICCV 2023

An early project on self-supervised image denoising using multi-view disentanglement.

Source

Contact

Email is the best way to reach me about research, software, or related work.

Chenyuan.Qu@outlook.com

Other addresses

Allsee
henry.qu@allsee-tech.com
Vieunite
henry.qu@vieunite.com
University of Birmingham
cxq134@student.bham.ac.uk