***Sensing the unseen through gaze, touch, writing, robotics, and generative systems***

From 8 to 13 September 2026, the Computational Media and Arts (CMA) Thrust at The Hong Kong University of Science and Technology (Guangzhou)  presents six artworks at the Ars Electronica Campus Exhibition in Linz, Austria. The exhibition will take place at Hörsaal E (H8.EG.04), Kunstuniversität Linz (University of Arts Linz), Hauptplatz 8, Ground Floor (EG), 4020 Linz, Austria. This marks the fourth consecutive year that CMA student works have been presented on this international media arts platform. (For info, please visit our event webpage: Click here)

Presented under the curatorial theme "Reality Flux: Sensing the Unseen," the exhibition brings together interactive installations, generative sculpture, robotic art, multi-channel moving image, and interactive textiles. Across the six works, acts of looking, touching, writing, whispering, and generating become ways of sensing forces that ordinarily remain beyond direct perception.

Since its founding in 1979, Ars Electronica has brought together artists, scientists, and technologists to examine the changing relationships among media art, digital technology, and society. In 2026, CMA enters this conversation through a reality that is contingent, relational, and continually in flux. Here, the unseen is not merely concealed; it is structurally inaccessible, emerging only through exchanges among bodies, machines, materials, and environments.

Within "Reality Flux: Sensing the Unseen," artificial intelligence, robots, algorithms, and sensors do not operate as autonomous authors. They serve as intermediaries within relational systems. A visitor's gaze, touch, gesture, voice, or movement can alter a work as it unfolds. Meaning is not fixed in advance; it is negotiated and co-produced in the moment of participation.

Artwork Highlights

1. "The Unseen Realm" - Jue Wang, supervised by Clea von Chamier-Waite

"The Unseen Realm" is an interactive projection installation driven by real-time eye tracking. A visitor's gaze, saccades, and blinks perturb a virtual cosmos of 64,000 particles while simultaneously reshaping generative cosmic textures projected across the installation surface.

The project began with an abandoned quarry in Dongguan, China. Through on-site scanning, image capture, and algorithmic reconstruction, the artist connects the geological traces of industrial activity with cosmic forces such as dark matter and dark energy. Once the visitor's eye enters the system, looking is no longer passive: eye and cosmos affect one another and jointly produce the work's continuously shifting reality.

2. "Touch Our Scars" - Yufan Li and Qing Yang; Yufan Li supervised by Kang Zhang

"Touch Our Scars" uses textiles, bodies, and an interactive system to explore trauma as a non-linguistic, embodied experience. The installation comprises two interactive textiles informed by collective trauma reports and nine touchable works developed from individual sensations of trauma.

Sharp, rigid, and physically resistant materials are woven into traditional textile structures. Visitors may approach, observe, linger, withdraw, or reach out to touch the surfaces. Rather than turning suffering into spectacle, the work asks audiences to confront the hesitation, responsibility, and potential transformation involved in touching another person's scars.

3. "Becoming Space" - Xinyu Ma and Hengyu Meng supervised by Clea von Chamier-Waite and Hungbo Fu

Comprising three 3D-printed resin sculptures, generative AI processes, and moving image, "Becoming Space" draws on the human-animal transformations of Ovid's Metamorphoses to reconsider identity, embodiment, and power under contemporary technological conditions.

Text prompts, three-dimensional generative models, algorithms, machines, and materials all participate in shaping the sculptural forms. Human and animal bodies no longer retain stable boundaries but pass into one another through an ongoing process of transformation. Generative AI is positioned not simply as a tool for depicting a predetermined image, but as a participant in an open-ended process of becoming.

4. "BruSHŪ" - Tiancheng Liu and Nicolò Merendino; Tiancheng Liu supervised by Kang Zhang

Chinese calligraphy is commonly encountered as a silent, finished mark, yet the approach, hesitation, turn, contact, lift, and re-entry that constitute the act of writing largely disappear from the final trace. "BruSHŪ" transforms a sensor-embedded brush into a cross-modal perceptual instrument, capturing orientation, motion energy, and the distance between brush tip and paper in real time.

These signals are translated into sound and evolving visual textures, allowing the tacit rhythms and decisions of writing to become perceptible. The work neither evaluates the quality of the calligraphy nor attempts to reconstruct the resulting marks. Instead, it turns the brush from a tool for producing traces into an interface for sensing embodied knowledge.

5. "I Tell the Moon My Secret and the Moon Tells Me Yours" - Zoe Qi-Jing Li, supervised by Clea von Chamier-Waite

In "I Tell the Moon My Secret and the Moon Tells Me Yours," visitors are invited to speak a secret into a microphone. Their voices are converted into waveforms, coordinates, and robotic movements. Using the Moon as its only light source, a robotic arm then traces these encoded secrets through long-exposure photography.

Even when the robot executes each trajectory with precision, the final image remains dependent on lunar phase, cloud cover, air quality, and whether the Moon enters the camera's field of view. The natural environment becomes an active and uncontrollable collaborator rather than a passive backdrop. Whether the secret appears or dissolves into the night, it remains secret - translated into another language.

6. "KARMA" - Hao Li, supervised by Pan Hui

"KARMA" is a multi-channel audiovisual installation that approaches generative artificial intelligence through the artist's framework of sensory projection. It imagines high-dimensional latent space as an "algorithmic cosmos" composed of vast datasets and innumerable possibilities. Human prompts, parameter settings, and artistic decisions intervene in this field, guiding latent possibilities into perceptible images and sounds.

Across multiple screens, real-world city footage, AI-generated imagery, text, detection boxes, and granular-synthesis soundscapes layer, dissolve, and recombine. The title points to cyclical causality and feedback within generative systems. The work asks how reality, aesthetics, and meaning are produced when human perception continually shapes algorithms, while algorithms in turn reorganize how reality is perceived.

From the Guangzhou Cross-Campus Exhibition to Linz

The six works selected for "Reality Flux: Sensing the Unseen" were previously presented as part of the 2026 Second Cross-Campus Exhibition in Guangzhou, "Symbiotic Signals: Entangled Perceptions and Algorithms." The exhibition connected artistic, design, and technological practices across HKUST and HKUST(GZ), providing a platform for cross-campus exchange, public presentation, and the continued development of student work.

As the projects move from Guangzhou to Linz, they are not being transferred as fixed outcomes. Each work continues to evolve through a new venue, new technical conditions, and new relationships with audiences. In this sense, the transition itself becomes part of the reality in flux that the exhibition sets out to examine.

CMA's Continuing Commitment to the Ars Electronica Campus Exhibition

Since 2023, the CMA Thrust at HKUST(GZ) has maintained an active presence at the Ars Electronica Campus Exhibition, treating the international platform as an important part of its support for student-led artistic practice, interdisciplinary research, and global exchange.

Guided by the strategic vision of CMA Thrust Acting Head Prof. Pan Hui and former Thrust Head Prof. Kang Zhang, "Reality Flux: Sensing the Unseen" is curated by CMA faculty members Clea von Chamier-Waite (lead), Ingeborg Reichle, and James She. The curatorial team will accompany the participating artists to Linz, supporting the presentation of their work and its engagement with an international audience.

CMA is one of four Thrusts within the Information Hub at HKUST(GZ). Its research and practice span augmented and virtual reality art, AI art, robotic art, sonic arts, games, design, data visualization, human-computer interaction, computer graphics, and moving image. By translating research methods into artistic experiences that can be publicly encountered and activated, CMA continues to explore how emerging technologies reshape relationships among people, bodies, society, and the environment.

Event Information

Exhibition: Reality Flux: Sensing the Unseen

Platform: Ars Electronica Campus Exhibition 2026

Dates: 8-13 September 2026

Venue: Hörsaal E (H8.EG.04)
Kunstuniversität Linz (University of Arts Linz)
Hauptplatz 8, Ground Floor (EG)
4020 Linz, Austria

Media, Partnership, and General Enquiries

Contact: Emily Liu    |    Email: cmat@hkust-gz.edu.cn

On June 17, the large-scale cross-regional MR×AI art exhibition SURREALITY, jointly presented by the Hong Kong University of Science and Technology (HKUST) and HKUST (Guangzhou), officially opened on the campus of HKUST (Guangzhou). "SURREALITY" is the world's first large-scale cross-regional MR×AI art exhibition, hosted by the Metaverse and Computational Creativity Research Center at HKUST (Guangzhou). It is one of the important events celebrating the 35th anniversary of HKUST, taking place simultaneously on the campuses of both HKUST and HKUST (Guangzhou).

"SURREALITY" integrates mixed reality (MR), virtual reality (VR), artificial intelligence (AI), spatial computing, and digital art creation, aiming to explore a new artistic form that embodies the coexistence of future virtual worlds and physical spaces. This exhibition received submissions from over 70 artists from 23 countries and regions around the world, showcasing approximately 50 works on the campuses of both HKUST and HKUST (Guangzhou). The participating artists and their works have been recognized with international awards such as the Emmy Awards, the immersive category at the Venice Film Festival, the Lumen Awards, and the South by Southwest (SXSW) Arts Festival. Their works have also been featured at prominent international art institutions, including the Museum of Modern Art in New York, Tate Modern in London, the Centre Pompidou in Paris, and the Karlsruhe Center for Art and Media in Germany. This exhibition not only presents the latest practices in cutting-edge technology and art but also showcases the collaborative efforts of HKUST and HKUST (Guangzhou) in interdisciplinary integration, computational creativity, and the cultivation of innovative talents in the Guangdong-Hong Kong-Macau Greater Bay Area.

、

Nansha, Guangzhou, September 2025 —The Computational Media and Arts (CMA) thrust at HKUST(GZ) will present five student artworks—four from CMA, HKUST(GZ) and one from HKUST—to the Ars Electronica Campus Exhibition in Linz, Austria (3–7 September 2025). Chosen from more than 20 submissions across both campuses, the pieces underwent a two-round jury led by Prof. James She and comprising CMA faculty, former Campus Exhibit student artworks, and external artists and experts.

Since 1979, Ars Electronica has gathered artists, scientists and visionaries in Linz each September to probe—and often provoke—the future through media art, digital experimentation and urgent debate on technology’s role in society. Under this year’s CMA curatorial theme “Phantom Protocols: When Machines Hallucinate in Panic”, the exhibit space in POSTCITY, Linz, Austria becomes a twilight theatre. Here, artificial intelligences murmur in half-sleep, immersive media swell and ebb like living biomes, and ecological thought is felt as a pulse within the walls. Five artworks—conjured by the thrust’s boldest master’s and doctoral alchemists—fracture the membrane between carbon and code, projecting prismatic after-images of a society yet to arrive. (Click here for our exhibit booklet)

Artworks Highlights:

1. "Drift of the Uncharted" — Ary-Yue Huang, Wenzong Ma and Bingyuan Wang, guidance by Clea von Chamier-Waite

Drift of the Uncharted is an immersive installation where a quadruped robot, evoking a robotic search and rescue agent in a speculative future world, is equipped with a projector that traverses the physical and virtual space, projecting scenes of sea-level rise onto the exhibition area, as if it were a remote agent of humanity, exploring the drowned cityscapes. This artwork critically meditate the hyperobject of climate change into an embodied reckoning, while re-scripting the robot from tool to storyteller.

2. "Algorithmic Miner" — Jia Sun and Zheng Wei, guidance by Pan Hui

Algorithmic Miner immerses participants in a virtual world where their actions—image labeling, sentiment annotation, and more—fuel the ascent of golden jellyfish toward a digital utopia. Using generative AI, symbolic metaphors, and spatial audio, the work critically examines the commodification of creativity and human agency in AI systems. It asks: Who benefits from the “invisible work” behind every intelligent machine, and what is the cost to our own autonomy?

3. "Capillary Network" — Hanlu Ma, Hengyu Meng, Robert Jankowski,  Ziwei Wu, Jindu Wang and Tianrui Hu, guidance by Huamin Qu and Zeyu Wang

Invisible toxins circulate through the intertwined ecologies of life, passing through species to species. Capillary Network visualizes this interspecies transmission as a living atlas: generative veins bloom across floor-to-ceiling screens, their pulses animated by real contamination data drawn from wildlife, insects, companion animals, and humans. Approach the work, and your silhouette merges into the flow—motion-tracking grafts your body into the network, revealing how toxins bypass the boundaries of form and category, weaving all living beings into a shared and fragile circulatory system. Conceived under the guidance of Huamin Qu and Zeyu Wang, the installation transforms ecological threat into an elegiac choreography of data, engaging speculative fabulation as both method and medium. Rather than framing data solely as evidence of loss, Capillary Network opens a space for imagining kinship beyond the human: kinship not limited to lineage or affinity, but arising through cohabitation, vulnerability, and the shared imperative to survive.

4. "Machine Civilization" — Yixiong Wang, Hourun Wang, Jiaqi Shi and Yufeng Zeng, guidance by Chen Liang and Tania Fraga

Machine Civilization reimagines the dawn of civilization: many robot observers collect images and data as the “first recorder,” feeding an AI parliament where artificial agents debate, negotiate, and evolve their own symbolic language. This generative process, visualized through dynamic scripts on projection surfaces, provokes audiences to consider what happens when machines gain linguistic sovereignty—and how humanity might coexist with truly autonomous digital civilizations.

5. "Chimpanzee, Run!" — Siyuan Liu, guidance by Tengjia Zuo

Set in a near-future world on the brink of extinction, Chimpanzee, Run! lets players navigate perilous landscapes as endangered chimps. Through immersive gameplay and narrative choices, the work spotlights humanity’s role in species loss, environmental destruction, and ethical decision-making. It compels participants to question whether change is possible—or if our desires will always outweigh the needs of our fellow beings.

CMA’s Enduring Initiative: The Ars Electronica Campus Exhibit Commitment

Building a long-term constellation at Ars Electronica since year 2023, the CMA thrust of HKUST(GZ) operates as both launchpad and laboratory for radical art-tech futures. Anchored by the strategic vision of CMA thrust Head Prof. Zhang Kang and Acting Head Prof. Pan Hui, and driven by Faculty-in-Charge Prof. James She, the initiative this year fields a supportive tribunal of expertise — core faculty (Professors Theo Papatheodorou, Clea von Chamier-Waite, David Yip, Jake Zhang, Luwen Yu, Tenjia Zuo, and more), battle-tested alumni from Campus Exhibit 2024 (CMA students Jarry Yang and Jiaqi Shi) and external luminaries (artists Jieyuan and Ying Ling)—to ensure every chosen work meets festival gold-standard.

For this year’s Ars Electronica Campus Exhibition, the expedition grows: Professors Clea von Chamier-Waite and Ingeborg Reichle will accompany Prof. James She and the student cohort, bringing curatorial vision and presentation expertise to amplify visibility, ignite global networks, and propel HKUST(GZ) artists onto the world stage.

For inquiries, partnerships, or media requests, please reach out to  
Prof. James She
Email: jamespmshe@hkust-gz.edu.cn

CMA is invited to be part of the Campus program at the Ars Electronica Festival 2025 in Linz, Vienna. Hence, this open call serves for selecting the best artwork from CMA and HKUST(GZ) students to showcase at the Ars Electronica festival 3-7 Sep 2025 in Linz, Austria.

Brief Background: A total of 42 international universities and 285 projects working in the academic fields of media art, design and technology participate in the Ars Electronica Festival 2024. Through this collective endeavor, the Campus Exhibition celebrates the power of creativity, collaboration and exploration in shaping our shared future.

References: Please find here some reports and images about the campus exhibition from 2024 in this page: https://ars.electronica.art/hope/en/campus/. You can also check two of our news (1) and (2) last year for more info.

Our Goal: 3-4 student art projects will be selected for exhibiting at Ars Electronica Campus Exhibit 2025 and at least one student artist of each selected project will have a chance to participate in the festival and represent CMA in Linz, given the funding supports / commitment from the student supervisor allows. Otherwise, the student must seek for the alternative funding supports from the University or other means by themselves.

Our Plan and Steps (Tentative):

#1 Registration: Contact Prof. James She (by Wechat msg.) to indicate your interests before Mar. 24, 2025 – 5pm (Mon.).

#2 Briefing: Attend a 20-30mins. briefing (in person ideally, or online) that is likely to happen on Mar. 26, 2025 (Wed.) early afternoon. Exact timing, venue and link will be notified.

#3 Proposal Preparation: Before the briefing, please prepare a bit the following proposal requirements:

- Short project description (150-300 words)

- Detailed project description, including the technical realization

- Images and/or sketches of proposed artwork

- Technical and special needs

- If applicable, video documentation of the project

- Short CV of each co-author (1-2 pages)

- Feasibility evaluation of demo and transportation.

- Funding planning

#4 Proposal Submission: Due Time (announced during briefing)

#5 Student Presentation and Jury Evaluation: A Committee of Jury will be formed from the CMA faculty members or/and invited experts to select projects based on the proposal, student presentation, funding condition, etc.

#6: Final Decision by Jury, and Final Actions by Students: Details later

All inquiries regarding this open call may be directed to Prof. James She over Wechat msg. in the Wechat group (AE participation).

Additional references:

1. Campus Award 2024

Computational Media and Arts (CMA), HKUST (GZ)

The Hong Kong University of Science and Technology (Guangzhou)

CMA PhD student Xuanyang (Nolan) Huang was named a finalist artist in the Arte Laguna Prize for his AI-generated artwork. The piece was exhibited at the Arsenale Nord in Venice, Italy, from November 17th through December 8th.

Arte Laguna Prize is a prestigious international art award based in Venice, Italy. Established in 2006, it promotes contemporary art and emerging artists worldwide through exhibitions, collaborations, and residencies. The prize covers various categories including painting, sculpture, digital art, and installations. Among the 10,000 international submissions, Xuanyang's AI artwork 'Imaginary Sunsets' was selected as one of the 200 finalists for the 23’24 edition Arte Laguna Prize exhibition at the Arsenale Nord.

'Imaginary Sunsets', Xuanyang's nominated work, is an AI-generated artwork that explores the connection between artificial intelligence and collective memory. Using clouds as a metaphor, the work incorporates manually collected images of clouds photographed through windows worldwide during the pandemic period as training datasets. The piece investigates how human collective memory during a specific period is mediated and interpreted through technology, and how machines poetically witness history.

Xuanyang was selected as one of six finalist artists in the digital art category.

The CMA Thrust 2024 PhD New Student Orientation on September 1, 2024, ended with a buzz of excitement and energy, as PhD students and faculty came together for a lively and warm start to their academic journey.

Kicking off the event, Thrust Head Professor ZhANG Kang welcomed everyone with open arms and enthusiasm, setting the tone for a day filled with warmth and high spirits. Introducing the new faculty members, he brought a fresh wave of energy and optimism into the academic community.

Associate Head of Teaching Professor Theo patiently and meticulously explained the curriculum and academic outline of the CMA PhD program to the new students, providing professional guidance on course selection to help students better plan their academic development paths.

PG Coordinator Professor Fan shared essential information on University regulations and key academic resources, ensuring that the new students are well-equipped to navigate their academic journey with confidence and support.

In the end, the Orientation event arranged a diverse and colorful tea break for the faculty and students, allowing everyone to exchange thoughts and experiences in a relaxed and enjoyable atmosphere. The on-site calligraphy fan activity also attracted many enthusiastic participants, allowing everyone to share and inspire each other in the joy of calligraphic art.

We extend our heartfelt thanks to all the new students for their enthusiastic participation. With a sense of camaraderie and excitement now firmly in place, we look forward to seeing new friendships bloom and academic horizons expand. Let's embark on this academic adventure together, ready to embrace challenges, celebrate achievements, and write our own unique and brilliant academic story. Here's to exploring the boundless sea of knowledge, growing together, and making unforgettable memories along the way. The journey ahead promises to be exciting, and we can't wait to experience it with all of you!

20 AUG 2024

Guangzhou, China - The Thrust of Computational Media & Arts (CMA) at Hong Kong University of Science & Technology (Guangzhou) is excited to share that five media and AI artworks will be showcased at the renowned media art event, Ars Electronica, in Linz, Austria. Led by Prof. James She, CMA, a group of talented student artists will be presenting their innovative and creative artworks at the Ars Electronica Campus Exhibit in Linz, Austria during September 4-8, 2024.

These carefully reviewed and selected artworks by a panel of CMA faculty members, external media artist and art educators, exploring themes of inclusivity, cultural connections, emotional healing, and hope for the future, represent a fusion of traditional art forms and cutting-edge technology. Co-curated by, Ms. Fiona Wang You (CMA Ph.D. candidate) and Prof. James She, the exhibition theme "Convergence: A Melodic Dance of Language, Ink, and Emotion in the Cosmos" showcases the harmonious blend of tradition and technology through AI, media technologies and cultural heritage.

Featuring artists such as Jiayang Huang, Yue Huang, Yulin Shen, Joshua Nijiati Alimujiang, Fiona You Wang, Hiimorit, Jiaqi Shi, Tianyu Lin, Zhen Wu, Changkun Liu, and Xingxing Yang, mainly from the CMA of HKUST Guangzhou campus, along with a group of student artists from the Hong Kong campus, each artwork offers a unique perspective and invites viewers to engage in a sensory experience that celebrates the fusion of Eastern and Western influences, heritage inclusivity, cosmic dialog through sound and senses, and emotional connections in Virtual Reality.

Highlighted artworks include:

1. "Ephemera" by Jiayang Huang, Yue Huang, and Yulin Shen: An interactive installation exploring linguistic taboos and the intersection of art and technology.

Ephemera is an interactive installation addressing linguistic taboos. Using a surreal “virus” as an AI embodiment of banned words, we employ generative models for visuals and large language models for communication. This project simulates a world of forbidden language, prompting introspection on unconscious discourse and AI's understanding of censored language. Our work, proposing Language as Virus, Art as Antibody offers new perspectives on language taboos and the art-technology intersection.

2. "Ink Sonata" by Jiaqi Shi and Tianyu Lin: Blending Eastern and Western cultures using AI to transform piano music into emotional calligraphic art.

Ink Sonata merges Eastern and Western cultures using AI to deconstruct piano music into emotions and reconstruct them as calligraphic art. Viewers can explore “calligraphic works” inspired by Beethoven and Pudsey, interact with the art through an interactive piano installation and immerse themselves in the emotional essence of music through oriental ink lines. Ink Sonata challenges cultural boundaries, redefining “understanding” in the digital era and showcasing technology's impact on art.

3. "Hum" by Fiona You Wang and Hiimorit: A blend of data from the James Webb space telescope with ancient nomadic chants, creating a sensory journey connecting viewers with the universe.

After growing up in inner Mongolia, I embraced the nomadic spirit and cosmic nomadism by living in various countries. By converting data from the James Webb space telescope into audible waves and blending them with ancient nomadic chants, viewers can experience a sensory journey that connects them with the universe through organic sounds. This exploration invites audiences to engage with the interplay of sound, senses and human experience, enabling them to actively participate in a cosmic dialog.

4. "DanceYipékda" by Joshua Nijiati Alimujiang: Integrating AI technology with traditional Uyghur Atlas fabric to preserve heritage inclusively.

DanceYipékda is a transformative project that integrates AI technology with traditional Uyghur Atlas fabric to preserve heritage inclusively. The exhibit features 3D-printed sculptures adorned with vibrant colors and intricate designs, inviting tactile engagement. Through dynamic projection mapping, the artwork comes to life, creating a captivating dialog between physical forms and digital enhancements while also fostering cultural connections and inclusivity in the modern age.

5. "Tame" by Zhen Wu, Changkun Liu, and Xingxing Yang: A musical journey embracing affective emotions and bonds in Virtual Reality.

VR, often seen as an escapist realm, offers solace detached from real-world constraints. What if we could feel genuine affection here, connecting with others on a deeper level? Inspired by The Little Prince, our Tame project is a multi-sensory journey exploring taming, bonding and attach ment. Players, as the Little Prince, traverse diverse locations, discovering emotional bonds. With generative AI for music, Tame enhances engagement and offers a healing, interactive VR experience.

The CMA Thrust at HKUST (Guangzhou) is dedicated to pushing the boundaries of creativity and technology for art, media and design. With their continued participation in Ars Electronica, the initiative receives the vision and support from the CMA Thrust head, Prof. Zhang Kang and the supporting faculty members and student artist supervisors, Prof. James She, Prof. Theo Papatheodorou, Prof. Masu Raul, Prof. David Yip, Prof. Zeyu Wang, Prof. Luwen Yu, Prof. Varvara Guljajeva, Prof. Tristan Braud (HKUST Hong Kong campus), etc., showcasing their supportive and persistent commitment to innovation in creativity for art, media and culture on a global scale.

For more information about the exhibit and the featured artworks, please check our exhibit brochure, or visit the websites of Ars Electronica and the Thrust of Computational Media & Arts at HKUST (Guangzhou)

Contact:

Prof. James She

Associate Professor, Computational Media and Arts, Information Hub, HKUST(GZ)

jamespmshe@hkust-gz.edu.cn

Assistant Professor Varvara Guljajeva opened this September a solo show ABELLÓ, IMAGINARY LANDSCAPE, at the Joan Abello Museum in Barcelona, Spain. This exhibition captivates tribute to Joan Abelló's centennial. Drawing inspiration from her summer residency in Abelló's archives in 2021, the exhibition offers a contemporary reinterpretation of his legacy. It features traditional and AI-generated portraits, merging past and future artistic styles, and transports visitors on imaginary journeys through landscapes that Abelló cherished, reimagined through generative AI. The exhibition culminates with 'Vision of Destruction', an interactive piece showcasing the transformative power of perspective. Varvara & Mar's innovative integration of technology in art echoes throughout the exhibit, inviting reflections on the intersection of art, history, and digital innovation.

Varvara Guljajeva

Assistant Professor

Computational Media and Arts (CMA), HKUST (GZ)CMA Info

From October 2 to October 6, the International Conference on Computer Vision (ICCV 2023) was held at the Paris Convention Center. A total of 10 results were accepted by the Information Hub of the Hong Kong University of Science and Technology (Guangzhou), among which a total of 5 results were accepted by Wang Lin.

PhD in Mechanical Engineering

Korea Advanced Institute of Science and Technology, 2021

Assistant Professor

Thrust of Computational Media and Arts

Faculty Profile | HKUST(GZ)

HKUST(GZ)

Dynamic PlenOctree for Adaptive Sampling Refinement in Explicit NeRF

Haotian Bai, Yiqi Lin, Yize Chen, Lin Wang

Abstract: The explicit neural radiance field (NeRF) has gained considerable interest for its efficient training and fast inference capabilities, making it a promising direction such as virtual reality and gaming. In particular, PlenOctree (POT) , an explicit hierarchical multi-scale octree representation, has emerged as a structural and influential framework. However, POT’s fixed structure for direct optimization is sub-optimal as the scene complexity evolves continuously with updates to cached color and density, necessitating refining the sampling distribution to capture signal complexity accordingly. To address this issue, we propose the dynamic PlenOctree (DOT), which adaptively refines the sample distribution to adjust to changing scene complexity. Specifically, DOT proposes a concise yet novel hierarchical feature fusion strategy during the iterative rendering process. Firstly, it identifies the regions of interest through training signals to ensure adaptive and efficient refinement. Next, rather than directly filtering out valueless nodes, DOT introduces the sampling and pruning operations for octrees to aggregate features, enabling rapid parameter learning. Compared with POT, our DOT outperforms it by enhancing visual quality, reducing over 55.15/68.84% parameters, and providing 1.7/1.9 times FPS for NeRF-synthetic and Tanks & Temples, respectively.

[1] Yu, Alex, et al. "Plenoctrees for real-time rendering of neural radiance fields." Proceedings of the IEEE/CVF International Conference on Computer Vision. 2021

Dynamic PlenOctree for Adaptive Sampling Refinement in Explicit NeRF

The explicit neural radiance field (NeRF) has gained considerable interest for its efficient training and fast inference capabilities, making it a promising direction such as virtual reality and gaming. In particular, PlenOctree (POT)[1], an explicit hierarchical multi-scale octree representation, h…

arXiv.orgHaotian Bai

OmniZoomer: Learning to Move and Zoom in on Sphere at High-Resolution

Zidong Cao, Hao Ai, Yan-Pei Cao, Ying Shan, Xiaohu Qie, Lin Wang

Abstract: Omnidirectional images (ODIs) have become increasingly popular, as their large field-of-view (FoV) can offer viewers the chance to freely choose the view directions in immersive environments such as virtual reality. The Mobius transformation is typically employed to further provide the opportunity for movement and zoom on ODIs, but applying it to the image level often results in blurry effect and aliasing problem. In this paper, we propose a novel deep learning-based approach, called OmniZoomer, to incorporate the Mobius transformation into the network for movement and zoom on ODIs. By learning various transformed feature maps under different conditions, the network is enhanced to handle the increasing edge curvatures, which alleviates the blurry effect. Moreover, to address the aliasing problem, we propose two key components. Firstly, to compensate for the lack of pixels for describing curves, we enhance the feature maps in the high-resolution (HR) space and calculate the transformed index map with a spatial index generation module. Secondly, considering that ODIs are inherently represented in the spherical space, we propose a spherical resampling module that combines the index map and HR feature maps to transform the feature maps for better spherical correlation. The transformed feature maps are decoded to output a zoomed ODI. Experiments show that our method can produce HR and high-quality ODIs with the flexibility to move and zoom in to the object of interest. Project page is available at http://vlislab22.github.io/OmniZoomer/

Cao_OmniZoomer_Learning_to_Move_and_Zoom_in_on_Sphere_at_ICCV_2023_paper.pdf

Look at the Neighbor: Distortion-aware Unsupervised Domain Adaptation for Panoramic Semantic Segmentation

Xu Zheng, Tianbo, Yunhao Luo, Lin Wang

Abstract: Endeavors have been recently made to transfer knowledge from the labeled pinhole image domain to the unlabeled panoramic image domain via Unsupervised Domain Adaptation (UDA). The aim is to tackle the domain gaps caused by the style disparities and distortion problem from the non-uniformly distributed pixels of equirectangular projection (ERP). Previous works typically focus on transferring knowledge based on geometric priors with specially designed multi-branch network architectures. As a result, considerable computational costs are induced, and meanwhile, their generalization abilities are profoundly hindered by the variation of distortion among pixels. In this paper, we find that the pixels’ neighborhood regions of the ERP indeed introduce less distortion. Intuitively, we propose a novel UDA framework that can effectively address the distortion problems for panoramic semantic segmentation. In comparison, our method is simpler, easier to implement, and more computationally efficient. Specifically, we propose distortion-aware attention (DA) capturing the neighboring pixel distribution without using any geometric constraints. Moreover, we propose a class-wise feature aggregation (CFA) module to iteratively update the feature representations with a memory bank. As such, the feature similarity between two domains can be consistently optimized. Extensive experiments show that our method achieves new state-of-the-art performance while remarkably reducing 80% parameters.

Zheng_Look_at_the_Neighbor_Distortion-aware_Unsupervised_Domain_Adaptation_for_Panoramic_ICCV_2023_paper.pdf

A Good Student is Cooperative and Reliable: CNN-Transformer Collaborative Learning for Semantic Segmentation

Jinjing Zhu, Yunhao Luo, Xu Zheng, Hao Wang, Lin Wang

Abstract: In this paper, we strive to answer the question ‘how to collaboratively learn convolutional neural network (CNN)- based and vision transformer (ViT)-based models by selecting and exchanging the reliable knowledge between them for semantic segmentation?’ Accordingly, we propose an online knowledge distillation (KD) framework that can simultaneously learn compact yet effective CNN-based and ViT-based models with two key technical breakthroughs to take full advantage of CNNs and ViT while compensating their limitations. Firstly, we propose heterogeneous feature distillation (HFD) to improve students’ consistency in low-layer feature space by mimicking heterogeneous features between CNNs and ViT. Secondly, to facilitate the two students to learn reliable knowledge from each other, we propose bidirectional selective distillation (BSD) that can dynamically transfer selective knowledge. This is achieved by 1) region-wise BSD determining the directions of knowledge transferred between the corresponding regions in the feature space and 2) pixel-wise BSD discerning which of the prediction knowledge to be transferred in the logit space. Extensive experiments on three benchmark datasets demonstrate that our proposed framework outperforms the state-of-the-art online distillation methods by a large margin, and shows its efficacy in learning collaboratively between ViT based and CNN-based models.

Zhu_A_Good_Student_is_Cooperative_and_Reliable_CNN-Transformer_Collaborative_Learning_ICCV_2023_paper.pdf

Benchmarking and Analyzing Robust Point Cloud Recognition: Bag of Tricks for Defending Adversarial Examples

Qiufan Ji, Lin Wang, Cong Shi, Shengshan Hu, Yingying Chen, Lichao Sun

Abstract: Deep Neural Networks (DNNs) for 3D point cloud recognition are vulnerable to adversarial examples, threatening their practical deployment. Despite the many research endeavors have been made to tackle this issue in recent years, the diversity of adversarial examples on 3D point clouds makes them more challenging to defend against than those on 2D images. For examples, attackers can generate adversarial examples by adding, shifting, or removing points. Consequently, existing defense strategies are hard to counter unseen point cloud adversarial examples. In this paper, we first establish a comprehensive, and rigorous point cloud adversarial robustness benchmark to evaluate adversarial robustness, which can provide a detailed understanding of the effects of the defense and attack methods. We then collect existing defense tricks in point cloud adversarial defenses and then perform extensive and systematic experiments to identify an effective combination of these tricks. Furthermore, we propose a hybrid training augmentation methods that consider various types of point cloud adversarial examples to adversarial training, significantly improving the adversarial robustness. By combining these tricks, we construct a more robust defense framework achieving an average accuracy of 83.45% against various attacks, demonstrating its capability to enabling robust learners. Our codebase are open-sourced on: https://github.com/ qiufan319/benchmark_pc_attack.git.

Ji_Benchmarking_and_Analyzing_Robust_Point_Cloud_Recognition_Bag_of_Tricks_ICCV_2023_paper.pdf